Merge pull request #1749 from wong00/fix-flashattn3-attention-compute

[bugfix] Fixed the logic of computing attention in flashAttention3.
2025-12-10 11:52:18 +02:00
parent dd58511d4e 38a48c670a
commit eb6dd96049
1 changed files with 1 additions and 1 deletions
@@ -66,7 +66,7 @@ else:
                    0, dtype=torch.int32).to(q.device, non_blocking=True),
                seqused_q=None, seqused_k=None, max_seqlen_q=lq, max_seqlen_k=lk,
                softmax_scale=softmax_scale, causal=causal,
-                deterministic=deterministic)[0].unflatten(0, (b, lq))
+                deterministic=deterministic).unflatten(0, (b, lq))
        else:
            assert FLASH_ATTN_2_AVAILABLE
            x = flash_attn.flash_attn_varlen_func(q=q, k=k, v=v,