A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 08:15 UTC
[BugFix] Fix type error when assign a trition kernel tensor to a torch.nn.Parameter (#28603)
Signed-off-by: Zijing Liu <liuzijing2014@gmail.com>
Zijing Liu committed
Nov 13, 2025 at 03:30 UTC
5e973209aaf5fa15459555eaa42bcd20ea63aa0d
1 file changed
+2
-2
vllm/model_executor/layers/quantization/mxfp4.py
+2
-2
@@ -755,8 +755,8 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
755
756
self.w13_weight = w13_weight
757
self.w2_weight = w2_weight
758
- layer.w13_weight = w13_weight
759
- layer.w2_weight = w2_weight
758
+ layer.w13_weight = Parameter(w13_weight.data, requires_grad=False)
759
+ layer.w2_weight = Parameter(w2_weight.data, requires_grad=False)
760
else:
761
raise ValueError(f"Unsupported backend: {self.mxfp4_backend}")
762