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

fix: add missing clamp_limit for CompressedTensorsWNA16MoE (#27328)

Browse Source
This commit is contained in:
Qeeweew
2026-06-16 19:29:57 -07:00
committed by GitHub
parent 9c53853ea3
commit 0ae4740bd1
1 changed files with 1 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py
+1
View File
@@ -425,6 +425,7 @@ class CompressedTensorsWNA16MoE(CompressedTensorsMoEScheme):
num_bits=self.num_bits, num_bits=self.num_bits,
is_k_full=self.is_k_full, is_k_full=self.is_k_full,
routed_scaling_factor=self.moe_runner_config.routed_scaling_factor, routed_scaling_factor=self.moe_runner_config.routed_scaling_factor,
clamp_limit=self.moe_runner_config.swiglu_limit,
workspace=layer.workspace, workspace=layer.workspace,
) )
return StandardCombineInput(hidden_states=output) return StandardCombineInput(hidden_states=output)