Update README.md

README.md

@@ -9,6 +9,7 @@ Updated. Was having some issues there with the hybrid attention and tensor shari
 
 Drop-in enhanced givens rotary block --  Its like a rubiks cube of embbedings :)
 
+Think of regular rope embeddings as a rotating 3d block in space.. Now add columns and rows that rotate and then rotate the faces of each resulting cube : 
 
 class CombinedRotaryEmbedding(nn.Module):
     def __init__(self, n_state, n_head, num_rotations, base=10000, checkpointing=False):