Fix skip layer in get_quant_method (#12632)

2025-11-04 23:27:46 +08:00
parent bb517fe393
commit 7cee07a067
1 changed files with 4 additions and 1 deletions
--- a/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
@@ -123,7 +123,10 @@ class CompressedTensorsConfig(QuantizationConfig):
        if should_ignore_layer(
            prefix, ignore=self.ignore, fused_mapping=self.packed_modules_mapping
        ):
-            return UnquantizedLinearMethod()
+            if isinstance(layer, LinearBase):
+                return UnquantizedLinearMethod()
+            return None
+
        if isinstance(layer, LinearBase):
            if CompressedTensorsConfig.DeepSeekFP8Config is not None:
                return Fp8LinearMethod(CompressedTensorsConfig.DeepSeekFP8Config)