xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-07 06:09:09 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix ViT with FlashAttention on ROCm (#30703)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni 2025-12-15 14:45:21 -05:00 committed by GitHub
parent ec154c36ee
commit 51e5b3e3c4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/attention/layer.py
View File

@ -464,7 +464,10 @@ class MultiHeadAttention(nn.Module):
} }
self.fa_version = None self.fa_version = None
if self.attn_backend == AttentionBackendEnum.FLASH_ATTN: if (
self.attn_backend == AttentionBackendEnum.FLASH_ATTN
and current_platform.is_cuda()
):
self.fa_version = get_flash_attn_version() self.fa_version = get_flash_attn_version()
assert self._flash_attn_varlen_func is not None assert self._flash_attn_varlen_func is not None
self._flash_attn_varlen_func = functools.partial( self._flash_attn_varlen_func = functools.partial(