[4/N] MoE Refactor: Unified Triton Kernel for FusedMoE and EPMoE (#8515)

2025-07-31 02:34:02 -07:00
parent e7dc163f57
commit 32fa1e9cc2
6 changed files with 70 additions and 690 deletions
--- a/python/sglang/srt/layers/quantization/unquant.py
+++ b/python/sglang/srt/layers/quantization/unquant.py
@@ -204,14 +204,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
        routed_scaling_factor: Optional[float] = None,
    ) -> torch.Tensor:

-        from sglang.srt.layers.moe.ep_moe.layer import EPMoE
-
-        if isinstance(layer, EPMoE):
-            return layer.run_moe(
-                hidden_states=x,
-                topk_output=topk_output,
-            )
-
        return self.forward(
            x=x,
            layer=layer,