Fix bug

llm_service/inferencer.py

@@ -206,16 +206,15 @@ def llm_inference(args):
     logger.info(f"Get params: model_path {model_path}, use_vllm {use_vllm}, tensor_parallel_size {tensor_parallel_size}, stream_chat {stream_chat}")
 
     model, tokenzier, sampling_params = init_model(model_path, use_vllm, tensor_parallel_size)
-    llm_infer = LLMInference(model,
-                            tokenzier,
-                            sampling_params,
-                            use_vllm=use_vllm,
-                            stream_chat=stream_chat)
 
     async def inference(request):
         start = time.time()
         input_json = await request.json()
-
+        llm_infer = LLMInference(model,
+                                tokenzier,
+                                sampling_params,
+                                use_vllm=use_vllm,
+                                stream_chat=stream_chat)
         prompt = input_json['query']
         history = input_json['history']
         if stream_chat: