Langchain-Chatchat Agent 不会调用工具

问题描述 / Problem Description Agent chat 的时候知道该用什么工具，但是并不会调用工具调用记录：

环境信息 / Environment Information

langchain-ChatGLM 版本/commit 号：2.10
使用的模型（ChatGLM2-6B / Qwen-7B 等）：ChatGLM3-6B
使用的 Embedding 模型（moka-ai/m3e-base 等）：bge-large-zh-v1.5
使用的向量库类型 (faiss / milvus / pg_vector 等)：pg_vector

我看有人提出这样的 issue，但是没有解决方法，有人有经验么

Apr 01 '24 09:04 xiaohundun

Hey @xiaohundun, long time no see! Hope you haven't been too swamped with those tricky issues. Let's crack this one together! 🛠️

Based on the information you've provided, it seems like the issue might be related to the tool invocation logic in your application. The StructuredChatOutputParserWithRetries class in the Langchain-Chatchat codebase is responsible for parsing the output from the language model and identifying tool calls. However, the actual invocation of the tool based on the parsed action is not shown in the provided code snippet.

Here are some steps you can take to troubleshoot and potentially resolve the issue:

Ensure that the tool invocation logic is correctly implemented and integrated within the agent's workflow. This typically involves mapping the action name to a specific tool or function call within your application, passing the extracted parameters to this tool or function, and handling the tool's response and integrating it back into the chat session. You can refer to the conceptual example provided in the context for more details.
Verify that the chatglm3-6b model and its tokenizer are correctly placed in the directory specified by MODEL_ROOT_PATH in your model_config.py.example file. Also, check the integrity of the chatglm3-6b model files, especially the tokenizer model file, to ensure they are not corrupted. Re-download or obtain a fresh copy if necessary.
Ensure that the custom tool is included in the tools sequence when initializing StructuredGLM3ChatAgent in the initialize_glm3_agent function. Also, check that the tool's name is correctly recognized in the parse method of the StructuredChatOutputParserWithRetries class.
Remove the OpenAI-related model configurations and only keep the chatglm configurations in model_config.py. Also, try changing the port number in both server_config.py and model_config.py from 8888 to 7777, and if that doesn't work, change it back to 8888.

If these steps do not resolve the issue, it may be necessary to provide more information or seek further assistance from the community or the maintainers of the Langchain-Chatchat repository.

Here are some relevant issues that might help you:

You can also refer to the source code of the StructuredChatOutputParserWithRetries and StructuredGLM3ChatAgent classes in the Langchain-Chatchat codebase for more insights:

server/agent/custom_agent/ChatGLM3Agent.py

I hope this helps! If you have any more questions or need further clarification, feel free to ask.

Sources

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.