From 388c05d54435939c3481e637093a4e10a0e19b7d Mon Sep 17 00:00:00 2001
From: yhyang201 <47235274+yhyang201@users.noreply.github.com>
Date: Fri, 19 Sep 2025 02:44:43 +0800
Subject: [PATCH] Fix bias handling in TritonMoeQuantInfo within
 quantization/mxfp4.py (#10579)

---
 python/sglang/srt/layers/quantization/mxfp4.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/python/sglang/srt/layers/quantization/mxfp4.py b/python/sglang/srt/layers/quantization/mxfp4.py
index 0d98d00d6..8643a3e36 100644
--- 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)