[Bugfix] Fix model run _npu_flash_attention hang issue (#4410)

Fix model run _npu_flash_attention in _forward_prefill_no_cache hang
issue, it was caused by wrong attention mask dtype.
### How was this patch tested?
Yes, tesed on Qwen2.5-VL and Qwen2.5-Omni

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

Signed-off-by: Ting FU <futing10@huawei.com>
This commit is contained in:
Ting FU
2025-11-29 09:20:22 +08:00
committed by GitHub
parent 048d350f9e
commit 9af34755ff
3 changed files with 6 additions and 7 deletions

View File

@@ -67,8 +67,6 @@ class AttentionMaskBuilder:
def get_attn_mask(self, max_seq_len: int, dtype: torch.dtype,
device: torch.device):
if max_seq_len == 2048:
return self.chunked_prefill_attn_mask.to(torch.bool)
self._update_attn_cache(max_seq_len, dtype)
return self.attn_mask_cache[:max_seq_len, :max_seq_len].contiguous(
).to(device, non_blocking=True)