From 66989a76425280fb099c77cf00073db0fd54e497 Mon Sep 17 00:00:00 2001
From: YanbingJiang <yanbing.jiang@intel.com>
Date: Thu, 11 Jun 2026 09:16:16 +0800
Subject: [PATCH] Fix gpt-oss-20b with mxfp4 support for Xeon (#27782)

---
 python/sglang/srt/layers/quantization/__init__.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/python/sglang/srt/layers/quantization/__init__.py b/python/sglang/srt/layers/quantization/__init__.py
index d104c8f4f..612a016d7 100644
--- a/python/sglang/srt/layers/quantization/__init__.py
+++ b/python/sglang/srt/layers/quantization/__init__.py
@@ -55,6 +55,7 @@ from sglang.srt.layers.quantization.w8a8_int8 import W8A8Int8Config
 from sglang.srt.platforms import current_platform
 from sglang.srt.utils import (
     cpu_has_amx_support,
+    is_cpu,
     is_cuda,
     is_hip,
     is_mps,
@@ -99,7 +100,7 @@ BASE_QUANTIZATION_METHODS: Dict[str, Type[QuantizationConfig]] = {
 }
 
 
-if is_cuda() or (_is_mxfp_supported and is_hip()):
+if is_cpu() or is_cuda() or (_is_mxfp_supported and is_hip()):
     BASE_QUANTIZATION_METHODS.update(
         {
             "mxfp4": Mxfp4Config,
@@ -130,6 +131,7 @@ CPU_QUANTIZATION_METHODS = {
     "compressed-tensors": CompressedTensorsConfig,
     "awq": AWQCPUConfig,
     "gptq": CPUGPTQConfig,
+    "mxfp4": Mxfp4Config,
 }
 
 QUANTIZATION_METHODS = {**BASE_QUANTIZATION_METHODS}