minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Fix skip layer in get_quant_method (#12632)

Browse Source
This commit is contained in:
Ke Bao
2025-11-04 23:27:46 +08:00
committed by GitHub
parent bb517fe393
commit 7cee07a067
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
+3
View File
@@ -123,7 +123,10 @@ class CompressedTensorsConfig(QuantizationConfig):
if should_ignore_layer( if should_ignore_layer(
prefix, ignore=self.ignore, fused_mapping=self.packed_modules_mapping prefix, ignore=self.ignore, fused_mapping=self.packed_modules_mapping
): ):
if isinstance(layer, LinearBase):
return UnquantizedLinearMethod() return UnquantizedLinearMethod()
return None
if isinstance(layer, LinearBase): if isinstance(layer, LinearBase):
if CompressedTensorsConfig.DeepSeekFP8Config is not None: if CompressedTensorsConfig.DeepSeekFP8Config is not None:
return Fp8LinearMethod(CompressedTensorsConfig.DeepSeekFP8Config) return Fp8LinearMethod(CompressedTensorsConfig.DeepSeekFP8Config)