[Frontend] API support for beam search for MQLLMEngine (#9117)

vllm/entrypoints/openai/serving_completion.py

@@ -9,6 +9,7 @@ from fastapi import Request
 
 from vllm.config import ModelConfig
 from vllm.engine.async_llm_engine import AsyncLLMEngine
+from vllm.engine.multiprocessing.client import MQLLMEngineClient
 from vllm.engine.protocol import EngineClient
 from vllm.entrypoints.logger import RequestLogger
 # yapf conflicts with isort for this block
@@ -150,15 +151,16 @@ class OpenAIServingCompletion(OpenAIServing):
                     log_tracing_disabled_warning()
 
                 if isinstance(sampling_params, BeamSearchParams):
-                    if not isinstance(self.engine_client, AsyncLLMEngine):
-                        raise ValueError(
-                            "Beam search in the API server is only supported"
-                            " with AsyncLLMEngine. please add "
-                            "`--disable-frontend-multiprocessing` to "
-                            "use beam search.")
+                    assert isinstance(self.engine_client,
+                                    (AsyncLLMEngine,
+                                    MQLLMEngineClient)), \
+                    "Beam search is only supported with" \
+                    "AsyncLLMEngine and MQLLMEngineClient."
                     generator = self.engine_client.beam_search(
-                        prompt_inputs["prompt_token_ids"], request_id_item,
-                        sampling_params)
+                        prompt_inputs["prompt_token_ids"],
+                        request_id_item,
+                        sampling_params,
+                    )
                 else:
                     generator = self.engine_client.generate(
                         {