remove attn_masks

vllm/attention/ops/paged_attn.py

@@ -135,8 +135,6 @@ class PagedAttention:
         blocksparse_vert_stride: int = 0,
         blocksparse_block_size: int = 64,
         blocksparse_head_sliding_step: int = 0,
-        attn_masks: Optional[torch.Tensor] = None,
-        attn_masks_stride: int = 0
     ) -> torch.Tensor:
         if blocksparse_vert_stride is not None and blocksparse_vert_stride > 1:
             # use blocksparse paged attention
@@ -182,7 +180,7 @@ class PagedAttention:
                 blocksparse_local_blocks,
                 blocksparse_vert_stride,
                 blocksparse_block_size,
-                blocksparse_head_sliding_step
+                blocksparse_head_sliding_step,
             )
         else:
             # Run PagedAttention V2.
@@ -220,7 +218,7 @@ class PagedAttention:
                 blocksparse_local_blocks,
                 blocksparse_vert_stride,
                 blocksparse_block_size,
-                blocksparse_head_sliding_step
+                blocksparse_head_sliding_step,
             )
         return output