Fix fused MoE when routed_scaling_factor is None (#8709)

2025-08-03 12:42:01 +08:00
parent d9def43dcd
commit 9f47d686e5
1 changed files with 3 additions and 1 deletions
--- 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):