[quantization] Properly ignore quantization for layers excluded in quant_config (#11205)

2025-10-07 14:06:05 -07:00
parent 420c99acfe
commit cd4b39a900
2 changed files with 8 additions and 10 deletions
--- a/python/sglang/srt/layers/quantization/quark/quark.py
+++ b/python/sglang/srt/layers/quantization/quark/quark.py
@@ -65,7 +65,9 @@ class QuarkConfig(QuantizationConfig):
        if should_ignore_layer(
            prefix, ignore=exclude_layers, fused_mapping=self.packed_modules_mapping
        ):
-            return UnquantizedLinearMethod()
+            if isinstance(layer, LinearBase):
+                return UnquantizedLinearMethod()
+            return None

        if isinstance(layer, LinearBase):
            scheme = self.get_scheme(layer=layer, layer_name=prefix)