xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:25:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Temp fix] Disable torch.compile for Qwen2.5 VL's VisionBlock temporarily. (#27760)

Browse Source 
Signed-off-by: Chenheli Hua <huachenheli@outlook.com>
Signed-off-by: Roger Wang <hey@rogerw.io>
Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Chenheli Hua 2025-10-29 16:17:48 -07:00 committed by GitHub
parent b5d90f7400
commit 48eb8eba58
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/model_executor/models/qwen2_5_vl.py
View File

@ -460,15 +460,17 @@ class Qwen2_5_VisionAttention(nn.Module):
return output return output
@support_torch_compile( # (FIXME): Enable this after dynamic slicing is fixed
dynamic_arg_dims={ # See https://github.com/vllm-project/vllm/pull/27760
"x": 0, # @support_torch_compile(
"cu_seqlens": 0, # dynamic_arg_dims={
"rotary_pos_emb": 0, # "x": 0,
"seqlens": 0, # "cu_seqlens": 0,
}, # "rotary_pos_emb": 0,
mark_unbacked_dims={"seqlens": 0}, # "seqlens": 0,
) # },
# mark_unbacked_dims={"seqlens": 0},
# )
class Qwen2_5_VisionBlock(nn.Module): class Qwen2_5_VisionBlock(nn.Module):
def __init__( def __init__(
self, self,