EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[bugfix] Fix typo in modelopt quant: 'FusedMoE' object has no attribute 'local_num_experts' (#8768)

Browse Source
This commit is contained in:
Trevor Morris
2025-08-04 11:08:08 -07:00
committed by GitHub
parent 2fa0462c39
commit 9bd4872a34
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/quantization/modelopt_quant.py
View File

@@ -776,7 +776,7 @@ class ModelOptNvFp4FusedMoEMethod(FusedMoEMethodBase):
# GEMM 1 # GEMM 1
w13_weight = ModelWeightParameter( w13_weight = ModelWeightParameter(
data=torch.empty( data=torch.empty(
layer.local_num_experts, layer.num_local_experts,
2 * intermediate_size_per_partition, 2 * intermediate_size_per_partition,
# 2 fp4 items are packed in the input dimension # 2 fp4 items are packed in the input dimension
hidden_size // 2, hidden_size // 2,