biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Fix type error when assign a trition kernel tensor to a torch.nn.Parameter (#28603)

Browse Source 
Signed-off-by: Zijing Liu <liuzijing2014@gmail.com>
This commit is contained in:
Zijing Liu
2025-11-13 03:30:04 -08:00
committed by GitHub
parent e63fd44560
commit 5e973209aa
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/quantization/mxfp4.py
View File

@@ -755,8 +755,8 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
self.w13_weight = w13_weight self.w13_weight = w13_weight
self.w2_weight = w2_weight self.w2_weight = w2_weight
layer.w13_weight = w13_weight layer.w13_weight = Parameter(w13_weight.data, requires_grad=False)
layer.w2_weight = w2_weight layer.w2_weight = Parameter(w2_weight.data, requires_grad=False)
else: else:
raise ValueError(f"Unsupported backend: {self.mxfp4_backend}") raise ValueError(f"Unsupported backend: {self.mxfp4_backend}")