Fix fused MoE when `routed_scaling_factor is None` (#8709)

9f47d686 · Liangsheng Yin · GitHub · d9def43d · 9f47d686
Unverified Commit 9f47d686 authored Aug 03, 2025 by Liangsheng Yin Committed by GitHub Aug 03, 2025
Hide whitespace changes
Inline Side-by-side

Showing with 3 additions and 1 deletion

python/sglang/srt/layers/moe/ep_moe/layer.py python/sglang/srt/layers/moe/ep_moe/layer.py +3 -1

No files found.
--- a/python/sglang/srt/layers/moe/ep_moe/layer.py
+++ b/python/sglang/srt/layers/moe/ep_moe/layer.py
@@ -280,7 +280,9 @@ class EPMoE(FusedMoE):
            m_max * self.start_expert_id,
            BLOCK_SIZE=512,
        )
-        return output * self.routed_scaling_factor
+        if self.routed_scaling_factor is not None:
+            output *= self.routed_scaling_factor
+        return output


 class DeepEPMoE(EPMoE):