docs/how_to/tool_choice/

[giscus[bot]]

docs/how_to/tool_choice/

This guide assumes familiarity with the following concepts:

https://python.langchain.com/docs/how_to/tool_choice/

[Ketan-K17]

I can't seem to get the point of this functionality, especially the 'tool_choice' attribute. Can't i just set a separate llm object and pass it only 1 tool to ensure it uses it?
Is there some application i'm missing here?

[brendenpetersen]

If you only pass it 1 tool, the LLM should be able to decide for itself (based on inputs) whether or not to call it--or even how many times to call it.

For example, if you have a single calculator tool and add it to an LLM and ask "What is Python?", it shouldn't call the tool. Or if you say, "What is 12345*6789, and also what is sin(100)?" it should call the tool twice in a single LLM call.

[deepracercve]

As I understand - it's also entirely optional. Even if asking "what is 123X456" LLM might choose not to use the multiply tool.

[faizrazadec]

tool_choice doesn't work with the gemini? How can one force the tool call if using gemini, I used the gemini with langchain init_chat_model.
This is my implementation:

    model = llm_conf.MODEL, 
    api_key=llm_conf.API_KEY,
    temperature=llm_conf.TEMPERATURE,
).bind_tools(tools, tool_choice="user_systems")```