zhangsibo1129

本地部署 modelscope llm 时，为什么只取message的第0条消息？

4

为什么只取第0条消息，而不是把所有消息整合成 raw_prompt ```python def _chat_no_stream(self, messages: List[Dict], stop: Optional[List[str]] = None, **kwargs) -> str: # The ModelScopeLLM only supports str inputs? prompt = messages[0]['content'] if self.custom_chat and self.model.chat: response =...

Add plan-and-execute agent

2

This PR implements a Plan-and-Execute Agent to address Issue #222, integrating into the existing architectural design. The Planner inherits from Assistant, and the Executor is implemented using ReactChat. How to...