Pass kwargs to chat_template.messages2prompt (#3982)

Author: shell-nlp

lmdeploy/serve/async_engine.py

@@ -625,7 +625,8 @@ async def _get_prompt_input(self,
                                                sequence_start,
                                                tools=tools,
                                                enable_thinking=enable_thinking,
-                                               reasoning_effort=reasoning_effort)
+                                               reasoning_effort=reasoning_effort,
+                                               **kwargs)
         if prompt is None:
             raise ValueError(
                 f'You are using base template to handle chat task. Please specify a `--chat-template` name chosen from `lmdeploy list` if you want to use OpenAI messages input.'  # noqa
@@ -733,7 +734,8 @@ async def generate(
                                                         adapter_name,
                                                         tools=tools,
                                                         reasoning_effort=reasoning_effort,
-                                                        enable_thinking=enable_thinking)
+                                                        enable_thinking=enable_thinking,
+                                                        **kwargs)
             prompt = prompt_input['prompt']
             input_ids = prompt_input['input_ids']
             self.request_logger.log_inputs(session_id=session_id,