Fix bias handling in TritonMoeQuantInfo within quantization/mxfp4.py (#10579)

2025-09-19 02:44:43 +08:00
parent fc809665fd
commit 388c05d544
1 changed files with 2 additions and 2 deletions
--- a/python/sglang/srt/layers/quantization/mxfp4.py
+++ b/python/sglang/srt/layers/quantization/mxfp4.py
@@ -731,8 +731,8 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
            quant_info = TritonMoeQuantInfo(
                w13_weight=layer.w13_weight,
                w2_weight=layer.w2_weight,
-                w13_weight_bias=layer.w13_weight_bias,
-                w2_weight_bias=layer.w2_weight_bias,
+                b13=getattr(layer, "w13_weight_bias", None),
+                b2=getattr(layer, "w2_weight_bias", None),
            )
            return self.runner.run(dispatch_output, quant_info)