update o1 results

ZeroEval-main/result_dirs/zebra-grid.summary.json

@@ -1,4 +1,15 @@
 [
+  {
+    "Model": "o1-mini-2024-09-12",
+    "Mode": "greedy",
+    "Puzzle Acc": "52.60",
+    "Cell Acc": "52.29",
+    "No answer": "0.80",
+    "Easy Puzzle Acc": "87.14",
+    "Hard Puzzle Acc": "39.17",
+    "Total Puzzles": 1000,
+    "Reason Lens": "993.28"
+  },
   {
     "Model": "claude-3-5-sonnet-20240620",
     "Mode": "greedy",
@@ -22,7 +33,7 @@
     "Reason Lens": "1153.83"
   },
   {
-    "Model": "Llama-3.1-405B-Instruct-Turbo",
+    "Model": "Llama-3.1-405B-Inst-fp8@together",
     "Mode": "greedy",
     "Puzzle Acc": "32.60",
     "Cell Acc": "45.80",
@@ -33,7 +44,7 @@
     "Reason Lens": "314.66"
   },
   {
-    "Model": "Llama-3.1-405B-Instruct-Turbo",
+    "Model": "Llama-3.1-405B-Inst-fp8@together",
     "Mode": "sampling",
     "Puzzle Acc": "32.60",
     "Cell Acc": "47.04",
@@ -76,6 +87,17 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1594.47"
   },
+  {
+    "Model": "Llama-3.1-405B-Inst@sambanova",
+    "Mode": "greedy",
+    "Puzzle Acc": "30.10",
+    "Cell Acc": "39.06",
+    "No answer": "24.70",
+    "Easy Puzzle Acc": "84.64",
+    "Hard Puzzle Acc": "8.89",
+    "Total Puzzles": 1000,
+    "Reason Lens": "2001.12"
+  },
   {
     "Model": "chatgpt-4o-latest-24-09-07",
     "Mode": "greedy",
@@ -164,6 +186,28 @@
     "Total Puzzles": 1000,
     "Reason Lens": "1389.75"
   },
+  {
+    "Model": "Llama-3.1-405B-Inst@hyperbolic",
+    "Mode": "greedy",
+    "Puzzle Acc": "25.00",
+    "Cell Acc": "46.62",
+    "No answer": "6.25",
+    "Easy Puzzle Acc": "66.67",
+    "Hard Puzzle Acc": "15.38",
+    "Total Puzzles": 16,
+    "Reason Lens": "1517.13"
+  },
+  {
+    "Model": "gemini-1.5-flash-exp-0827",
+    "Mode": "greedy",
+    "Puzzle Acc": "25.00",
+    "Cell Acc": "43.56",
+    "No answer": "8.50",
+    "Easy Puzzle Acc": "70.71",
+    "Hard Puzzle Acc": "7.22",
+    "Total Puzzles": 1000,
+    "Reason Lens": "1705.11"
+  },
   {
     "Model": "Meta-Llama-3.1-70B-Instruct",
     "Mode": "greedy",
@@ -176,7 +220,7 @@
     "Reason Lens": "1483.68"
   },
   {
-    "Model": "deepseek-chat",
+    "Model": "deepseek-v2-chat-0628",
     "Mode": "greedy",
     "Puzzle Acc": "22.70",
     "Cell Acc": "42.46",
@@ -209,7 +253,7 @@
     "Reason Lens": "1813.82"
   },
   {
-    "Model": "deepseek-coder",
+    "Model": "deepseek-v2-coder-0614",
     "Mode": "greedy",
     "Puzzle Acc": "21.10",
     "Cell Acc": "41.58",
@@ -220,7 +264,7 @@
     "Reason Lens": "1324.55"
   },
   {
-    "Model": "DeepSeek-Coder-V2-0724",
+    "Model": "deepseek-v2-coder-0724",
     "Mode": "greedy",
     "Puzzle Acc": "20.50",
     "Cell Acc": "42.35",
@@ -352,7 +396,7 @@
     "Reason Lens": "391.19"
   },
   {
-    "Model": "gemma-2-27b-it@nvidia",
+    "Model": "gemma-2-27b-it",
     "Mode": "greedy",
     "Puzzle Acc": "16.30",
     "Cell Acc": "41.18",
@@ -407,7 +451,7 @@
     "Reason Lens": "1043.90"
   },
   {
-    "Model": "gemma-2-9b-it@nvidia",
+    "Model": "gemma-2-9b-it",
     "Mode": "greedy",
     "Puzzle Acc": "12.80",
     "Cell Acc": "36.79",