[2/2] Refactor DeepGeem requant for FP8 FusedMoE on Blackwell (#13960)

python/sglang/srt/server_args.py

@@ -1517,6 +1517,11 @@ class ServerArgs:
                 self.ep_size == 1
             ), "FP8 Cutlass MoE is only supported with ep_size == 1"
 
+        if self.moe_runner_backend == "deep_gemm":
+            assert (
+                self.ep_size > 1 and self.moe_a2a_backend == "deepep"
+            ), "DeepGemm MoE runner is only supported when ep is enabled and moe_a2a_backend is deepep"
+
     def _handle_a2a_moe(self):
         if self.moe_a2a_backend == "deepep":
             if self.deepep_mode == "normal":