Expert Parallelism (EP) Support for DeepSeek V2 (#12583)

vllm/model_executor/layers/quantization/awq_marlin.py

@@ -464,10 +464,17 @@ class AWQMoEMethod(FusedMoEMethodBase):
         use_grouped_topk: bool = False,
         topk_group: Optional[int] = None,
         num_expert_group: Optional[int] = None,
+        global_num_experts: int = -1,
+        expert_map: Optional[torch.Tensor] = None,
         custom_routing_function: Optional[Callable] = None,
         scoring_func: str = "softmax",
         e_score_correction_bias: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
+        if expert_map is not None:
+            raise NotImplementedError(
+                "Expert Parallelism is not supported for "
+                "fused Marlin MoE method.")
+
         topk_weights, topk_ids = FusedMoE.select_experts(
             hidden_states=x,
             router_logits=router_logits,