minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[NPU]revert add scoring func for GLM 4.7 Flash (#31388)

Browse Source
This commit is contained in:
McZyWu
2026-07-16 09:16:57 +08:00
committed by GitHub
parent edb2059139
commit 871c648203
2 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/hardware_backend/npu/moe/topk.py
+1 -1
View File
@@ -97,7 +97,7 @@ def fused_topk_npu(
group_select_mode=(1 if use_grouped_topk else 0), group_select_mode=(1 if use_grouped_topk else 0),
renorm=0, renorm=0,
# 1 for sigmoid, 0 for softmax # 1 for sigmoid, 0 for softmax
norm_type=(0 if topk_config.scoring_func == "softmax" else 1), norm_type=1,
routed_scaling_factor=( routed_scaling_factor=(
topk_config.routed_scaling_factor topk_config.routed_scaling_factor
if topk_config.apply_routed_scaling_factor_on_output if topk_config.apply_routed_scaling_factor_on_output
python/sglang/srt/models/glm4_moe_lite.py
-1
View File
@@ -225,7 +225,6 @@ class Glm4MoeLiteSparseMoeBlock(nn.Module):
quant_config=quant_config, quant_config=quant_config,
routed_scaling_factor=self.routed_scaling_factor, routed_scaling_factor=self.routed_scaling_factor,
prefix=add_prefix("experts", prefix), prefix=add_prefix("experts", prefix),
**({"scoring_func": "sigmoid"}),
) )
self.topk = TopK( self.topk = TopK(