{ "measurement": { "model.layers.0.self_attn": [ { "accuracy": 0.9495436047252855, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9533736768521761, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9639359022441664, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9710959635282818, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.969502263947537, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9737770573088997, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9706712333779586, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9756533974095395, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9831381032341405, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9833519490141618, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9877627656648034, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991355739926037, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885440727597788, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918629260439622, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991294275773199, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954162306691471, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9915650659485867, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972526789887955, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978175394629177, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.0.block_sparse_moe": [ { "accuracy": 0.9692538662960655, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9711697164334749, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9710782233037447, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9714081726576153, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943978680591834, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944986265740896, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948025702645904, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980221184852877, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982760744659525, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981509429451666, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985128466628099, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987345631969603, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99888242898803, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990258492333325, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991796945074671, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992393457183713, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992993007365026, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.self_attn": [ { "accuracy": 0.9894618313563498, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990940123796463, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919539886085611, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948936104774475, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995165938217389, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953536344201941, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968014394766406, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9969586875093611, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9969100295320937, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970294997880333, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997569306508491, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977564715633267, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997811890158214, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980065016762206, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987503742112925, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989503787732438, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990934270777201, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994360989351806, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997317829463435, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.block_sparse_moe": [ { "accuracy": 0.9560204211034273, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9605542170374017, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.965637555247859, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9656405197946649, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911527186632156, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996257698457492, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962864259355947, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978897510782668, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981350773259213, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972506457645642, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979560379134981, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987139629298135, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988933864392733, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992351353560623, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991406963059777, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991430549049064, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991806937185558, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.self_attn": [ { "accuracy": 0.9982463323364132, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984067286315718, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998600460785, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990893784714373, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991345984959289, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991754758122721, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994136602372715, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994612319983149, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995280280709267, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995459050038144, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995747700725731, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996053078211844, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996263765680947, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996575882266227, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997841124422848, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998267597004183, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998388810843033, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999910000288908, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999549106486436, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.block_sparse_moe": [ { "accuracy": 0.997742616620503, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9978197311100206, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9980484376612463, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981007407370367, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990165309098206, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990847358774197, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991717392480687, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994987957179546, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995293451080981, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999505682652326, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995448253558654, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999729363541854, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997828604261342, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998425544043513, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9998608165494117, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998770622585557, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999940776737946, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.self_attn": [ { "accuracy": 0.9979736848097098, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981793278926298, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984487611800432, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998887591750214, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990158795722222, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990502908629807, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99929187133124, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993451698438117, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994271829920379, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994485499337316, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995180280192902, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995552409244212, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99958071676328, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996193368478041, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997569095392368, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998077953258824, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998092390474325, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998977769619638, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999447138489861, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.block_sparse_moe": [ { "accuracy": 0.9933877618689286, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9936925710031861, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9937198373832201, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937773931183314, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968125218231427, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922916387256823, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923232574211923, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970293448943841, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924108880130869, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987398144838057, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987730399558419, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987734903238321, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990891078977209, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988237997811091, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9996298394460035, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996525927103663, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997047520928869, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.self_attn": [ { "accuracy": 0.9952749162912369, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971197192606173, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974069330645235, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979667393000502, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981681882942978, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982264316395709, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989959568177399, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999051065037125, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991571112094741, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991803024160234, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992006188748699, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992288477032593, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993579099444967, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994245197035765, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999632155865823, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997048516953854, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997245390703412, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998508007510712, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999266876103846, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.block_sparse_moe": [ { "accuracy": 0.9964204556063602, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9965415349916408, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9968878124889574, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996965474792217, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983333433537107, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984634493134523, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998581010162046, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991399925202131, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992089727403302, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991644909115214, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992683738175976, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995711012498328, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996304390461821, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997633350277809, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997826152432122, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998044559094859, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999368877493238, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.self_attn": [ { "accuracy": 0.996462465508988, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9967786207010871, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972943747906309, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979218338665209, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982101736884368, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982881893060709, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988585438971457, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989128862752726, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990248750699194, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990658376758036, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999098403477355, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991763147868609, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992922200380188, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993623114240012, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995969169969229, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996706342912818, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996922380339942, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998304634722636, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999128355044186, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.block_sparse_moe": [ { "accuracy": 0.9956583784599053, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9958059442670721, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.996236201179655, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963346582494284, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979444295167923, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981134924057283, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982583791409668, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989617519864911, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990497019730116, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989608706612336, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990946259349585, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994744480538525, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995484524161408, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997184378045955, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999730134431861, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997564684599638, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999239362629229, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.self_attn": [ { "accuracy": 0.9961234731109518, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963572060591296, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969633774537789, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975789103068804, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981131873240596, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981560569844747, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988172291532943, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988450489350056, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989093971487722, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989447202533484, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990209446925866, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990961535980827, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992399225991807, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993043262511492, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995736852288246, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996440202046774, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996805535629392, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999808126680651, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999905859659377, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.block_sparse_moe": [ { "accuracy": 0.9943205098572531, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9945239287457968, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9951216202033194, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952550883355894, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972746360458826, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975452570146636, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977467579669074, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985862729188643, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987549410250626, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986401118926311, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988195493229126, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993003629344074, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992749637090846, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996061462565864, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9996422370895743, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996789018378446, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998872077190563, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.self_attn": [ { "accuracy": 0.9955010123943028, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959623445021478, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964473337719315, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997305093431159, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997851914088977, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978849946668273, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985626686952616, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986222879470963, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987350655229468, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987975315828073, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998929384409597, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989994940789122, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990804043451422, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991449214714138, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994872962369731, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995666272252014, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999622680295847, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999763341498022, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998903189981846, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.block_sparse_moe": [ { "accuracy": 0.9932020141890174, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9934422005163995, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9941864680302771, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943535339675451, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9967985847278645, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970515835049906, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997302317501683, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983879532665014, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985181333399132, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998386813523738, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985891606070494, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991810485034397, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992983580419892, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995635123923421, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995792559102962, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996261898869354, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998837260308823, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.self_attn": [ { "accuracy": 0.9944688689551855, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9951258525252342, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957726993843129, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969208783617145, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973824418296939, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974461399803036, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983053203476103, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984060771959392, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985245101545986, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985907559136027, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986939412590704, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987853108660171, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988825819210002, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989740546596678, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993839082063029, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994820821540136, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995637773781231, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997223338945523, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998742941227791, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.block_sparse_moe": [ { "accuracy": 0.992175721024212, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9924342742091731, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9932709148055628, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934583890594935, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963746666908264, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966587839942229, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969518931680604, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981391273046795, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982980711287573, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998167730083591, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983395234142479, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990597561393914, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999189185664842, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994823623350576, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995210555038954, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995757722246804, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998522210670145, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.self_attn": [ { "accuracy": 0.9938019765050787, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943699605370823, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951505253189489, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964264384225795, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969638442914737, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970512070546025, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980617969443923, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981475472450256, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983504471418104, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984455802722981, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984545324389872, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986042418566189, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987180926101772, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988442035882097, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992755614711266, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994103461994153, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994906387046764, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996884703930271, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998564204005035, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.block_sparse_moe": [ { "accuracy": 0.9908048643877632, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9911158602488669, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9921316261354246, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923618671141172, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9956526207296472, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959976677047578, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963485037809924, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977998664896739, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979755435335008, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997809190695223, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980812433518862, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988855016662886, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990432306535935, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994017951876709, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994267066940665, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994935506072483, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998334264137635, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.self_attn": [ { "accuracy": 0.993387833237648, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937570463669928, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944588220433185, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958049207925797, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966236517617577, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967310050208318, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978155097679088, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979340659552499, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980771818051213, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981907553186542, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983200909275758, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984527248889208, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998563602762787, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986847875345695, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991999754780218, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993298243927328, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994227571020785, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996374701639932, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999840070628316, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.block_sparse_moe": [ { "accuracy": 0.9901394389177623, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9904769259063821, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.991623571828792, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918822355960545, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953243093271005, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957009522538436, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960937629404821, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976340704842618, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978202779434229, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976531341671944, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979404714938841, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988081167010885, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989768495096972, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993642089202216, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9993876620244823, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994624513446501, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998285023642606, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.self_attn": [ { "accuracy": 0.9920875524219713, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992707117607719, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934222619784506, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950585428037142, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996216330481203, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962623350714382, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976267722484312, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976782430159418, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978635083687933, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997966547153498, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980840169285473, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982102244700256, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983286498800704, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998447329884297, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990636225005514, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992184007148209, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993570422949759, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995658109944902, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998147676968457, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.block_sparse_moe": [ { "accuracy": 0.988763083752833, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9891456640080402, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9904370480462125, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907275226555372, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9946469997889117, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950732314272931, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955081845584669, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973086171636456, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975242203003482, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973024474947076, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997637870672502, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998630395747329, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988258140847871, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992705851205086, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992966117141279, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993782634601781, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998071586075974, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.self_attn": [ { "accuracy": 0.9896929436608365, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905754484628376, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9915764810223329, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937431769151437, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950246026641444, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950941989296361, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968526547676638, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996928804995198, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971780110346643, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9973473103815004, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9974873897276426, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976619825159249, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997851084525648, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980145712432108, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998803679880343, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990008442025435, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991715798448575, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994198697079953, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997683739485709, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.block_sparse_moe": [ { "accuracy": 0.9870900519584355, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9875165475042242, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9890588278833189, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9894106427305623, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938403748367962, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943309057700006, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948552278311629, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969012911774611, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971528949314042, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968950064166596, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972825052314683, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998423956236557, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986490050428792, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991612264788464, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991907689132189, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999290062947885, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997778816491758, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.self_attn": [ { "accuracy": 0.989252295149, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899658717607197, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908918856005919, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932966295041536, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994593937538172, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946550044574236, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967132336214969, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967741272167155, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970724351311985, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972230315600571, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997274965832108, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974866062402725, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976705026469732, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997857140083062, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987031969388849, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989157960210976, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991455568294776, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994019257198823, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997637027776555, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.block_sparse_moe": [ { "accuracy": 0.9856689509592558, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9861679594767722, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9878821992560437, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882662547262091, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932241047683515, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937627370420256, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943431451132423, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965837911555642, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968621895501488, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965825959255821, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970114997735149, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982625287222234, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985134654531354, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990728925913572, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991069698804304, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992163933225369, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997537123088381, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.self_attn": [ { "accuracy": 0.9875651056829252, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9880292768541136, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892786145210266, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919710959258833, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936426764256076, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937460242133391, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962466864993698, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9963623053933445, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965741461829135, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967713795210186, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9968083642031017, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970155883776514, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972741939127445, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975022128538081, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985129258742458, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998741983485065, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990286184964996, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993155584425518, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997276817938607, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.block_sparse_moe": [ { "accuracy": 0.9842027739474648, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9847514676420313, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9866634011268616, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9870928745520743, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925095956576498, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9931051417400962, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937550280439226, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9962139843325866, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996522342688159, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9962159885387671, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966909818743405, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998075435232175, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983539126421276, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989713097089216, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990103755538401, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991331287708721, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997268968359813, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.self_attn": [ { "accuracy": 0.9861704707145691, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9866944175017508, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879216732163179, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907664617425517, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930063831178766, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930563098505923, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958791858271548, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959938079118729, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9962735325098038, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965158606830397, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965018437881219, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967407899859705, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969820791953489, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972429536283016, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983541383162925, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986192466397035, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989304204324359, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992387404567317, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997029240292153, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.block_sparse_moe": [ { "accuracy": 0.9824151553605732, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9830396175384521, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9851725477921335, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9856504452855963, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991680633080633, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.992340760011422, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930648458631415, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9957929430039305, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9961357108856502, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99580020849642, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963280119394001, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978655015951708, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981744930540261, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988582075426453, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989038495052802, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990389747054953, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996972017382321, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.self_attn": [ { "accuracy": 0.9840929774861586, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9846943240416678, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9860256078996157, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891214864818674, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919091282706511, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920216422331961, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952670214207548, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953793899009102, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957350366993954, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959728894265074, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959712193200463, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962855891177529, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965538355080705, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968176272354627, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981098482875448, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983970395436412, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987937360021629, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991085305809975, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996646673810717, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.block_sparse_moe": [ { "accuracy": 0.9807810940240559, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9814681881352475, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9838832977570986, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844244624439039, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9908932345478159, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9916269418440367, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924413467708387, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953910841753608, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9957733703287024, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9954097039605442, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959860134281611, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976687752886823, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980019442737103, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987513895489668, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988026315052259, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998953165486455, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996681931477628, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.self_attn": [ { "accuracy": 0.9817581929658589, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9824093470447942, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9840935123594183, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988109866255208, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907508429728056, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908468339004015, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947713095890848, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948341720982602, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952056878491452, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954873170507582, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953328371047974, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956428267453846, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960268041805217, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963387941059313, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978243838015356, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981669028730769, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99863829728412, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990182694439825, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996290208869859, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.block_sparse_moe": [ { "accuracy": 0.9791674127704219, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9799047316375532, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9824800475647575, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9830561405734012, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901442872850519, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9909183171234632, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9917885347416526, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950078808163342, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9954157938298426, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950327386981562, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956510580684009, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974796752396383, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978378752344533, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986500153808218, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987076624835792, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998871699171631, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999641090680502, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.self_attn": [ { "accuracy": 0.9797678730989757, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9805087870673129, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9823775542409796, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866095797011727, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896457838384729, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897361178147165, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939441904425621, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941153165541197, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945646006810037, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949170492197338, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947755305390609, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951788783073425, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955877920514659, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995938845763081, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975512935534904, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979629155836607, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984339245251919, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988934245744818, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995760586308805, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.block_sparse_moe": [ { "accuracy": 0.9779406111491354, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9787232169979497, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9814115025495228, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.982017750802793, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895594386677993, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.990381723956058, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912859334757453, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994695395231247, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9951276136072058, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947376741390479, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953910457460504, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973287094188364, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977086145234736, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985631423953333, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998631521079101, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988022129002371, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996172297304791, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.self_attn": [ { "accuracy": 0.9778726383259422, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9786646962165833, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.980605412470667, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9845187946369773, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884980211132451, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886526630113, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931916335695669, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932493583152169, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937866577192357, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943009583573592, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943173370863262, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946394774474596, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951621839090398, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99559370310683, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973501063099033, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977839994979533, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982413214288259, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987692128082639, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995263723754569, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.block_sparse_moe": [ { "accuracy": 0.9763915099595722, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9772229727945829, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9801384414497175, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980795725395805, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9887639495887255, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989657146366019, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9906302980686489, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942975252082473, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947611862107327, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943302105131903, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950378690111009, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971190863534024, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975306742677563, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984540876589323, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985230873878065, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987054752480042, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995901674326313, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.self_attn": [ { "accuracy": 0.9766747167235926, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.977801429597955, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.980324000120163, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848424070759824, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882929976049223, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884632496457351, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931054562330246, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993280153525503, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936303476753988, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940867024032694, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941543466166446, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945279634312579, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950703931482214, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954656127252077, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973180411677611, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997724059577051, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982101750609121, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987095549310508, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995164728086245, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.block_sparse_moe": [ { "accuracy": 0.974145034426137, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9750502940855528, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9782658627158717, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.978996887018806, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876832318933386, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9886612672554819, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9897328364221674, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937309261999632, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942362445749735, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937866757574835, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945590629389411, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968448928312251, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972930309411726, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982994653676686, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983864707970306, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985878310705486, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995462763564367, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.self_attn": [ { "accuracy": 0.9734739378878945, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9745850437565854, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9771044113134083, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9822285300806949, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863527040732535, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865098579933769, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918178053278672, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992011477288447, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926258545172842, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9930489314229864, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931956051211608, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935947864463455, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941141754388809, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946665167808533, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967860061871378, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973316086750281, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997860045417359, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985044909347045, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994280307406658, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.block_sparse_moe": [ { "accuracy": 0.9715773651474401, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9725644933549982, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9760104449171769, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9768016542259016, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9864875784045771, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9875548528997522, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9887015152918665, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99312256982452, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936789140889519, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9931865828601938, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940287247300148, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965385841695886, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970267540530154, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981382400973847, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998228318793209, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984456622286847, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995033755212238, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.self_attn": [ { "accuracy": 0.970573582147297, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9717523110540289, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9747428611705178, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9811433490953947, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9849702135512703, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851652807311008, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912774476565813, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914628769222059, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992064726980109, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9925406261494285, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924170759163404, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929962644451543, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936940979800726, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941753474505324, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965288929249111, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970868394563073, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977204346735227, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984082152184687, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993907099001502, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.block_sparse_moe": [ { "accuracy": 0.9708738703476755, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9719030417894062, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.975497211280622, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9763132804318478, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9862251705244968, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9873089311938537, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9884746866790872, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9930130407998436, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935654476284981, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9930516097106432, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9939122027472446, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964711924916819, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969709733206975, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998103640973568, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981929591219676, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984134769949474, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994953752642399, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.self_attn": [ { "accuracy": 0.9762766549461767, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9771520464043868, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9794852953208121, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9842026610123483, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9877105755241293, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879209720774701, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924686064845637, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926374562476811, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933870097524241, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937616880787047, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938353170689783, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942684412786835, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947909779454532, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952284144727808, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971435497465887, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997591924510504, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980563423351237, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986681067629865, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994642090444502, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.block_sparse_moe": [ { "accuracy": 0.9706553095265439, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9716470367030093, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9751286083146146, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9759302264765689, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9859276053152586, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9870614754526239, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.988212307032786, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9928810980759168, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934462120658473, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9928901540605646, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937840084496298, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963875558031233, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969086623505542, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99806339607427, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981490452038614, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983627629515371, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994794444524144, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.self_attn": [ { "accuracy": 0.9760159128590634, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9769509889577564, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9788438928754706, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9833660157103288, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875842089715757, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879263153201655, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928394297235891, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931087846818724, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935389663043775, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939415282324741, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938478720815558, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943537006252691, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948546623713092, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953055044538096, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971945091689888, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976525973332556, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981495767439666, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986923444820078, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999491609485918, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.block_sparse_moe": [ { "accuracy": 0.9691219016125328, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9701536987957201, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9736262497149015, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9744067160706771, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9851387425472862, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9863346250433671, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9874656208251652, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924949038969842, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9930920216598009, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924834700007188, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9934239701220864, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961775639339497, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967275514806572, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979584220992891, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980385346632255, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982480263631595, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994573118165135, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.self_attn": [ { "accuracy": 0.9771529640022077, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9792362435867912, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9814357679141196, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9856588918911783, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9887418017575615, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889885398902392, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930582313161147, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933440253922814, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939500924788023, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994285817209043, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944026976039535, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948914254966535, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953224953067931, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956987845270258, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973854250029514, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978247084899953, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982108603182592, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987768196549854, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999497890717497, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.block_sparse_moe": [ { "accuracy": 0.967997394110027, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9690795854518288, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9726157596236781, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9734087994224147, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845993361975017, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9858318488848837, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9869882393824426, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922176364221071, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99283193992941, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922216487558264, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9931854739000923, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960479230472916, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966087404050326, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978883848770669, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979739794997793, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981898857574714, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994366451313621, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.self_attn": [ { "accuracy": 0.9751289455514205, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9761519165415513, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9791339620163566, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9835000932216644, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876377715876228, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879233711644223, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924584446769011, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99271758920268, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934045523405075, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937784479636895, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939005257267701, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943624123146659, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948425896857914, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952586897109684, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971087253407428, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976103492081165, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980377401960524, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986637492517108, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994593022979403, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.block_sparse_moe": [ { "accuracy": 0.9659585764533595, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9670862273166054, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.970831971419485, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9716735262619821, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9836343040591792, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9849232482282739, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9861576698328319, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917222528081191, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923817617328543, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917335031848205, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.992748496563811, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958034506753871, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963934739169321, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977613375767281, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978495259818277, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980848537463891, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994052412282479, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.self_attn": [ { "accuracy": 0.9769132843143061, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9778640787852438, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.979995481277767, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9838078570993323, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879716446525172, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882152237390217, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925624610562074, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926767521782925, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931780719443372, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937365768771422, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937759814293761, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943701036666569, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948219127560917, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952996620222142, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971504291813624, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976137051064717, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980399487049956, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986645444051215, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994571285910512, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.block_sparse_moe": [ { "accuracy": 0.9648189230969078, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9659518411284999, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9698222436402973, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9707130601531581, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9830691924220637, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9844049340800235, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9857181342024552, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914195655207885, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921073497910249, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914569470443224, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925046665103812, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956631566348829, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962714228190874, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976813922587194, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977812067066368, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980298372867861, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993792425743059, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.self_attn": [ { "accuracy": 0.9731513450020238, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9743333788294541, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.977323885026731, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820983409881592, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9860554337501526, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865279448659796, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991123651987628, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9917963617726376, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924390825786089, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9928665349358007, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993159042377221, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936366524351271, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99424567504933, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947027615026424, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968294513068701, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973118520881, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977929941133449, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984883679763267, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993823263794184, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.block_sparse_moe": [ { "accuracy": 0.9635079126608999, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9646959994968615, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9689320231738844, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.969947127919448, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9823877811431885, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9837903427450281, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.985283291653583, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910183760680651, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917694024349514, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911160970989027, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922128555021787, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954919775849894, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961248781335982, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975754977448991, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997696162642617, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979790537373016, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993490451260617, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.self_attn": [ { "accuracy": 0.9713522666379025, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9726695826179103, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9759140751863781, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9814566671848297, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9853929579257965, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9857106538195359, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910413811081334, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9913363048904821, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9919549124805551, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9922879565703241, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926796618260836, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993226851287641, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939193255022952, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943697287848121, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966497291859827, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997176461706036, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976550864153787, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998395201014845, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993633538680641, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.block_sparse_moe": [ { "accuracy": 0.9639799751733479, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9651682878795423, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9692079958162809, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9701837646333795, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.982626924389287, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9839753006633959, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9854027785752949, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911525578875291, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918770343065262, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912270166371998, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.992303972181521, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955495380257305, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961706846952438, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976138031404269, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977248034587032, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979928880929947, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993587566147509, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.self_attn": [ { "accuracy": 0.9773769676685333, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787575555475134, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9811424612998962, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9853027227677797, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885289559238836, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9887974732800534, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927537300084767, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931203534728602, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936050058979737, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938919908906284, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942407545290495, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946309414349104, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951188399603492, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955083120026087, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973148742788717, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977393340515462, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981117536754984, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987585690656775, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994697088963891, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.block_sparse_moe": [ { "accuracy": 0.9623733633442929, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9635615662524575, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9677478796557376, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9687584513112119, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9817573020332738, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9831950601778532, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9846675662617934, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907314879329581, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914902221215399, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907914839292827, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9919244176463077, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953291690663287, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959839786353865, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974994818238836, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976118199135128, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978856497297162, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993290431111267, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.self_attn": [ { "accuracy": 0.9756264686584473, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9768787732249812, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9798085799342707, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9840762301495201, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875345810463554, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.987769218651872, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920815025505266, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923166436584372, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9928883376874422, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932299444549962, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938342657528425, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943066274649218, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947984708767188, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952342455324373, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971255507124098, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975560581997821, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979236147513515, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998673651171358, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999412367659572, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.block_sparse_moe": [ { "accuracy": 0.9610632469779566, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9622881067426581, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9664559301577116, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9674468730625353, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9810974080311624, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9825833424141532, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9840298891067505, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9904045815530577, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911911558163794, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990454666708645, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9916306365477411, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951608275112352, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958386381990031, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974152408540249, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975270048568123, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977990267308134, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993093756860808, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.self_attn": [ { "accuracy": 0.9762542279143083, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9779430831733503, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9805792789710196, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850002135101118, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882246718594903, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883939749316165, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992717948399092, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9929702117254859, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934410613618399, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937794314403283, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940888238580603, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945077456926045, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950477182865143, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995444880896493, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972528146677896, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977008247453916, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980833181425145, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986962576053644, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994658761118588, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.block_sparse_moe": [ { "accuracy": 0.9599959975794742, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9612322386942411, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9653653941656414, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9663582977495695, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9805281209318262, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9820627319185358, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9834925548026436, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901315467922311, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909372133644003, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901705101916665, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9913786273253592, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950198801724535, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957142354626405, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973452083374325, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974552794898811, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977196556957144, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992917016344635, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.self_attn": [ { "accuracy": 0.9759314562145033, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9774615419538397, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9798227454486647, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9837557344060195, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.987901193530936, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881640194278014, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923547059297562, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992638255420484, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993148115120436, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993477207657538, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939998074581748, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943799780387628, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948882472358251, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952791893952772, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971472368036446, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975983555379667, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979928535850424, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998620764793534, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994276218900555, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.block_sparse_moe": [ { "accuracy": 0.9589574713456004, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9602255978082356, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9644134515210202, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9654050343915036, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9800082727482444, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9815902584477475, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9830187480700644, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898748868390134, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990706930035039, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899027567160757, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991148347917356, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948783712951761, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956010513399777, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972735094396692, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973805201681036, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976411477515572, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992782921579323, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.self_attn": [ { "accuracy": 0.9749273642113334, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9763700977752083, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.978855267951363, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9837734024775656, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9869520734799536, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98711869199025, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923573442195591, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926473066995019, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932793104334882, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936431669875195, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936110294963184, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940749238196173, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948023738045442, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952293505009852, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971343563183358, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976028896083957, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99802326489436, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998655444305194, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994526909860341, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.block_sparse_moe": [ { "accuracy": 0.9594770795420596, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9607133488906057, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.964739485790855, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9656985119769448, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9802139307323255, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9817742830828616, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9831492006778717, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899841845035553, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9908075222843572, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990008007538946, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912389759954653, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949365207239201, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956448889876667, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973049106958666, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974092935260973, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976587875893241, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992826770696985, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.self_attn": [ { "accuracy": 0.982873830356096, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9836527642450834, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864373944307628, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9888101361299816, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910278359526082, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912319998992117, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994685024023056, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948808746902567, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952276368674479, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954087102883741, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953784852435714, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956627017573306, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996544249355793, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967997699583832, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981106072664261, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983877092599869, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986158327052468, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991494249552488, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995904195514557, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.block_sparse_moe": [ { "accuracy": 0.9577281067245885, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9590302235201785, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.963180422782898, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9641657126577277, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.979333432097184, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9809790881056535, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9823834378468362, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895436410841189, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9904036067034069, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895672609931544, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9908498909912611, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947083082638288, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954534267124376, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971856408213314, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972944534138629, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975519011679449, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992528595030308, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.self_attn": [ { "accuracy": 0.9795865385155929, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9803174075327421, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.982393165952281, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9860661970941644, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893421028789721, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989539802858704, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935601839893743, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938019968961415, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994228652825481, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994538737362937, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945710199443918, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949215144703263, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955160711940966, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958885963025846, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975388801019442, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979319866550597, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983095444346729, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988455720441906, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995137518761974, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.block_sparse_moe": [ { "accuracy": 0.9561480848412764, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9575962267423931, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9617322495109156, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9627194028151662, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787632383798298, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9804434854733316, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9818617419192666, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9892480412596151, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901324267450132, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9892915557873877, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9906104520747536, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945696639387231, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953324622229526, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971110154139368, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972239903321392, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974829785917935, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999232714456555, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.self_attn": [ { "accuracy": 0.9743566418948927, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.975552045985272, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9776418162019629, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9828236887329503, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9867861082679347, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.987088489689325, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918950788284603, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9922225898817966, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927467829302737, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931483017770868, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932795731644881, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937804604047223, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943460543688974, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948118572172365, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968505062555012, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973768942842358, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978376083860272, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985037616601116, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993890492049488, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.block_sparse_moe": [ { "accuracy": 0.9558051950053165, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9572206327789708, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9613725354796961, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9623665088101437, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9783616975734108, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9801128977223447, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9815235890840229, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890548030012533, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989963854614057, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890864060113305, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904350410950812, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944673635457691, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952433034777641, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970568142047054, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971710263114226, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997427010614621, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999216616202734, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.self_attn": [ { "accuracy": 0.9816641368364033, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825444833228463, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9852866326507769, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9878652229120857, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904010146856308, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908813849875802, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942848196155146, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945649202716978, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950073655498656, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952105440591511, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995227627848324, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955654991300482, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963577730875266, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966757697494406, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980004749407894, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998334093705604, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985004054676545, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990707741756188, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995725873465601, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.block_sparse_moe": [ { "accuracy": 0.9560552615868418, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9573947975510045, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9614514269326863, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9624187130677073, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.978374067105745, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9801176143319983, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9814738505765012, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989046081116325, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899505251332333, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890807521970648, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904265129252484, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944604599946424, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952421141298193, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970485838620287, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971699436244211, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974158786629376, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992156448332887, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.self_attn": [ { "accuracy": 0.9787498179234957, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9799091659094158, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9827480943579423, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864999228402188, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9890306890010834, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892999349456084, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933785725581018, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936552682989522, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941160874931436, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943921687571626, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944707170913094, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948524173937345, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956306336741698, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959746097263537, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975908712336892, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979715523751158, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982607286227377, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998848018873679, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994992933383113, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.block_sparse_moe": [ { "accuracy": 0.9549184159228676, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9562801405003196, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9603915779214156, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9613749039800543, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9777956071652865, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9795775099804527, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9809577245461313, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988754363436448, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9896864961636694, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9887808333886298, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9901664460960188, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943141631389919, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951112035073733, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969760895167527, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970934422392594, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973403135020482, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991975658034024, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.self_attn": [ { "accuracy": 0.9803921407774875, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9815161808540946, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9838100828622517, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9870139187888095, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989875603663294, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.990184735310705, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936408369164718, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939673358672544, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943947027388372, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946091347619107, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948485603458003, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952129615764869, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958732818302355, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962053306792912, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977187749586607, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980888307878846, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983121588041908, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988959669870766, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995191862904712, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.block_sparse_moe": [ { "accuracy": 0.9539741748257687, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9553724464617277, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9594849065730446, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9604718057732833, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9772969641183552, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9791209133047807, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9804993460052892, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988508057437445, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9894558622648841, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885254000362597, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9899479895830154, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9941823458985278, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950032634170431, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99690489725847, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970238200928035, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972667039224976, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999176296650579, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.self_attn": [ { "accuracy": 0.9772212928847263, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9784788034464184, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9812458659473219, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851836938607065, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885083327167913, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886837436964637, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930870815327293, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933364085460964, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937822003113596, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940505078748653, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941724774084593, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945824067843588, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952992749841589, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956690739644202, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973912733165842, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978044852614403, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998159957951621, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987471247778127, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994619524988689, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.block_sparse_moe": [ { "accuracy": 0.9529130490202653, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9543513624291671, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9584999398181313, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9594925987093073, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.976753747776935, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.978627943678906, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9800119086315757, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882260592360246, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9892078669447648, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882570199276272, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9897063625486273, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940461092873624, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948844925353402, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968354005954767, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969561488220566, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972010572489939, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991627717880827, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.self_attn": [ { "accuracy": 0.9767812333608928, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9779961360128302, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9807586277786055, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844945480949, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881845524436549, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885187533340956, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926608745989046, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9930492082708761, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935684576630592, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939004351433954, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939962260817227, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944408402631157, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950946004767167, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954840011502567, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972836853642213, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977400044077321, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980645534631453, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987036045640707, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994552763375012, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.block_sparse_moe": [ { "accuracy": 0.9519471024212084, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9533900869520087, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9575474952396593, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9585508359105963, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9762291359274011, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9781523001821417, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.979537612513492, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9879559874534607, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889711972914244, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9879920780658722, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9894812812930659, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9939138952054476, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947713535082968, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967634032823538, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968867411738948, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971310125761911, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991437901595706, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.self_attn": [ { "accuracy": 0.9846091380244807, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9852631594005384, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876380476512407, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897642096406535, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919882408882442, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922976540891748, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950436973258069, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953440482679167, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956741505547574, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995844203782709, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959898838087132, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962588082018652, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968550838530064, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971482477297908, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982469070114588, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998553452915267, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99867876109324, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991779775407753, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996140873010614, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.block_sparse_moe": [ { "accuracy": 0.9506573771175585, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9521644084077132, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9563484850682711, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9573596998264915, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9755708449765256, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9775526209881431, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9789412507885381, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876254310733393, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9886555616792879, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876590904436613, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9891838149020546, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937462955713272, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946270185081583, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966764603006212, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968026674499637, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970463700592518, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991226635481182, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.self_attn": [ { "accuracy": 0.9778145739906713, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9790969177296287, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9817713765721572, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9846952039944498, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886265679409629, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886160563481482, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927004277706146, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9928897203583467, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935206339547509, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938666977380451, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941129307997855, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994523755421764, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953305556585914, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958242903414526, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974596480789938, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997888520359993, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980912069348913, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998730062360042, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994275071903279, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.block_sparse_moe": [ { "accuracy": 0.9502306298205727, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9517495475317302, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9559643080360011, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9569743307013261, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9753531327373103, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.977346470481471, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9787262003672751, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9875147248569288, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885533601045609, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9875542254824388, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9890923625544498, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.993691016576792, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945805084548498, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966461117330351, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9967762584749021, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970178811957962, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991175843108642, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.self_attn": [ { "accuracy": 0.9769285713371477, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9779292705811953, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9806469035776038, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844272246486262, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882150794330397, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884249371917624, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927729344681689, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931030022470575, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935741471616846, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939097405264252, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939798312751871, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944393062277844, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950974646367525, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995495798948564, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972659376891035, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997727663305245, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980709872355586, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987084523431564, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994531995371768, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.block_sparse_moe": [ { "accuracy": 0.9480278774311668, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9496362962220845, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9539576266941271, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9549963223306757, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.974248324569903, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9763427125780206, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.977759439694254, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9869490752094671, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9880380505009702, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9869830929919293, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.988602307281996, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9934038731612658, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943352914170215, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964925595804265, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996627585276177, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968753933514419, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990756222488064, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.self_attn": [ { "accuracy": 0.9788503489996258, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9799576856588063, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9825684553698489, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861505737430171, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889559596776962, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893647000977868, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933485514239261, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936594374869999, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941098392009735, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943966214594088, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944884828046748, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948781760115373, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995549887418747, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959199495221439, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975246465519855, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979146950339016, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982459033398252, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998836299306468, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994902029437455, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.block_sparse_moe": [ { "accuracy": 0.9472590433923822, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9488672519984999, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.953190455311223, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9542298191472104, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9738268664008692, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9759470287122225, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9773602062150052, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9867345408389443, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9878425849111456, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9867815202788303, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9884160655109506, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9932999265821356, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9942444915834227, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964399412274361, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965772005288225, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968245598046404, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990609894065481, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.self_attn": [ { "accuracy": 0.9798208255516855, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806277344101354, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.983418848953749, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.986568232900218, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895284277828116, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9898080025848589, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938096894245398, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941510432644894, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946455473178312, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948856987451252, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994768274849967, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995109815738703, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958338419857778, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961678287700603, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976668393141345, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980484017808187, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983800398676019, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989755551673865, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995317659585884, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.block_sparse_moe": [ { "accuracy": 0.9449577707993357, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9466600041640432, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9510949599115472, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9521666106424833, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9726938699421129, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9749087612879904, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9763474809496027, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.986164529072611, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9873180201179103, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9862148369613447, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.987917390308882, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930170845044287, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9939970907412077, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962882638761872, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9964313522765511, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966830234778555, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990255487592596, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.self_attn": [ { "accuracy": 0.9751241865910982, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9762966303448928, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9793050728346172, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9840284460469296, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9869332627246254, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9872611941475617, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992277520267587, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9925370679089898, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931116676644275, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934816391844499, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935032380254645, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939412486396337, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994721487948769, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951517523119324, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970774507444156, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975625164806843, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979554128489996, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986526596506959, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994227148494438, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.block_sparse_moe": [ { "accuracy": 0.9420379149286371, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9438607190784655, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9484065331910786, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9495118479979665, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9712439461758262, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9735912178692064, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.975076509149451, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9853996995248293, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9866171369427129, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9854757268177835, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.987273644459875, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.992639809846878, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9936757515135565, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960751984464494, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9962372873958788, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964990909946593, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989651690580343, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.self_attn": [ { "accuracy": 0.9740156923469744, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9755663291404122, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9790815114974976, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9831780430517698, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866033980720922, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868759952093425, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918321378921208, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9921988680174476, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927549260227304, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931219706409856, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933498721373709, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937407099886945, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946793945212113, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951643586943024, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997089191486961, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975855856349594, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978663131202522, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986066856470547, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993957432575131, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.block_sparse_moe": [ { "accuracy": 0.9383435939487658, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9403775930404663, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9451173481188322, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.946289338563618, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9694720663522419, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9719936126156857, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9735715514735171, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844769615876047, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.98578973349772, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845886999054959, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9864932267289412, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921802776424509, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9932849305240732, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958235786149376, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960048426138727, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962886791480215, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988922697344893, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.self_attn": [ { "accuracy": 0.966326544159337, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9682306364962929, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9724030745656866, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9779224678089744, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9830024430626317, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9832630314325032, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900824208008615, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9904117803824576, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9910070958890413, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9915309618962439, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914807096907967, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920628549237, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931304439118034, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937432067174661, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962307328456327, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968200400471687, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973838462641365, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982407685173186, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992798905035383, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.block_sparse_moe": [ { "accuracy": 0.9376858912016216, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9395701446031269, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9440901091224269, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9452401023162038, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9696752899571469, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9721648222521732, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9737654642054909, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845412546082547, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9858309406983224, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9846903681755066, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986581846287376, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922232071035787, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933068360152998, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958118008155572, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960169929422831, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99631160066316, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988448887475228, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.self_attn": [ { "accuracy": 0.9610932343884518, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9632926959740489, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9684340451893053, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9748094034822363, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9807653803574411, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9811892493775016, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889862082506481, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9892172201683647, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9898943830477563, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9903875691326041, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9903162953100706, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908671598685416, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920727971353029, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926694953127911, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956607002961008, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963181203133181, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970960826858094, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979984891649923, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991902640873664, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.block_sparse_moe": [ { "accuracy": 0.9389861508419639, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9409818963000649, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9456299104188618, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9468022584915161, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9698010432092767, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9722670944113481, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9738492557876989, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845759288260811, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9858443517433969, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.984744037452497, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9866276508883426, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922626583199752, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933493678507052, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958391687587688, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960499852895737, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963408139975447, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988979819769922, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.self_attn": [ { "accuracy": 0.9623561376019528, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9644075694837069, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.968421901527204, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9748275201571616, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.981529585624996, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9817984794315538, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9890508259597578, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9893949008301685, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9901140159682223, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.990709082076424, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9907174024142718, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911966598347614, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992110926853983, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928277111367175, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956762359330529, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963427448743268, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971176087856293, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979842895347821, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991981661633441, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.block_sparse_moe": [ { "accuracy": 0.9394109186373258, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9414147389562506, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9458926790638974, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9470474657259489, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9699085103838068, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9723806255742123, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9739578899584318, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845656187910783, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9858244798685375, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9847451950374403, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986650581422605, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922501617356351, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933516892947649, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995801644889932, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960465427292021, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963428146745029, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988632608009013, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.self_attn": [ { "accuracy": 0.9565950098790621, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9590474429883455, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9650815631213941, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.972474029189662, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9783248273949874, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.978822083849656, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873944779759959, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9877801082636181, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9887372663146571, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9892986706997219, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9891828659333681, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989999602499761, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912921927477184, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919463772522775, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951688070830546, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995965755299518, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966926821752599, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978022036191664, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990899056000145, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.block_sparse_moe": [ { "accuracy": 0.9364966279581973, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9381859114295558, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9432707334819593, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9445803290919254, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.968742245122006, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9712397863990382, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9730697995737979, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9839527481480649, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9853132492617557, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9841804316169337, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9861272604841935, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9919660099242863, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930857294484189, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956688084884694, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.995880453210128, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962342918703431, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988038598706848, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.self_attn": [ { "accuracy": 0.9566341199372944, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9596392135871084, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9659288933402613, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9725109903435958, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9785895661303872, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9790050952058089, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875168361161885, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9878042050098118, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9889731062086005, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9895716952650171, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9893954309977984, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900851132054078, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914728544260326, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921594536618182, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952880461749277, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960456643449632, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968308282918051, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978790465546282, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991189394342271, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.block_sparse_moe": [ { "accuracy": 0.939212861813997, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9411778700979132, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9457957305406269, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9470242575595254, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9698360530953658, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9723183449945951, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9740192341177087, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844382687618858, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9857286010917864, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9846782135336023, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9865712288178896, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921707981511166, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9932907537410134, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957233385035866, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960062194027399, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963469222972268, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988539392422688, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.self_attn": [ { "accuracy": 0.9640823132113406, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9649236704173841, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9709980080002233, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9761721178105003, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9818046736089807, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9825476658971686, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900718143111781, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9902640510546533, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911111781471654, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914629639763581, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9915828908744612, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992036698680175, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930065493834647, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936265867007407, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961390942335129, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967300995792213, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974023580159012, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982267763269576, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999229824856708, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.block_sparse_moe": [ { "accuracy": 0.9519420642601817, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9535729571392662, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9572240835741946, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9583123106705516, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.977016839541887, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9789001549545088, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9803536314713327, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9877864326301374, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889839260201705, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988344710908438, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9897958001023844, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99399904318546, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948528036475182, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966262790717577, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969032710712207, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972357555831733, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989282826059743, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.norm.norm": null, "lm_head.linear": null }, "last_module_idx": 114 }