From 52d4c697bb462e54543d651b2277a24f935698ca Mon Sep 17 00:00:00 2001 From: Polisetty V R K Jyothendra Varma Date: Wed, 13 May 2026 01:35:51 +0530 Subject: [PATCH] Fix fused_moe import for non-NPU devices (#25076) Co-authored-by: Brayden Zhong --- python/sglang/srt/models/afmoe.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/python/sglang/srt/models/afmoe.py b/python/sglang/srt/models/afmoe.py index 543c8bb12..efdfaade7 100644 --- a/python/sglang/srt/models/afmoe.py +++ b/python/sglang/srt/models/afmoe.py @@ -62,9 +62,7 @@ from sglang.srt.utils import add_prefix, is_npu _is_npu = is_npu() -if not _is_npu: - from sglang.srt.layers.moe.fused_moe_triton import fused_moe -else: +if _is_npu: from sglang.srt.hardware_backend.npu.quantization.fused_moe_method_npu import ( fused_moe_npu as fused_moe, )