Update flash_dmattn/integrations/modeling_flash_dynamic_mask_attention_utils.py

Co-authored-by: Copilot <[email protected]>

Author: LoserCheems

flash_dmattn/integrations/modeling_flash_dynamic_mask_attention_utils.py

@@ -182,7 +182,7 @@ def _get_unpad_data(attention_mask: torch.Tensor) -> tuple[torch.Tensor, torch.T
     """
     seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
     indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
-    # NOTE: Similar to the `.item()` in prepare_fa2_from_position_ids, with torch compile,
+    # NOTE: Similar to the `.item()` in prepare_fdma_kwargs_from_position_ids, with torch compile,
     # this might cause a graph break
     max_seqlen_in_batch = seqlens_in_batch.max().item()
     cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))