docs/how_to/tool_calling/

[giscus[bot]]

docs/how_to/tool_calling/

This guide assumes familiarity with the following concepts:

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

[Carolmelon]

useful doc, ^_^

[lamoboos223]

amazing docs!

why tool calling isn't supported on llama3? i think it is the most important model among them all since it is open source and everyone can fine tune to there dataset.

[Xx-AD16-xX]

Try using llama3 from groq and use it via chatgroq instead of using it locally; I have tried and it works!

[vivekatbitm]

How can we pass methods of a class object, annotated as @tool. It's not working as first param of class instance methods is self. It complains while calling such tool method using .invoke call. For example: db_instance.get_students.invoke({}) fails saying
TypeError: StudentDB.get_students() missing 1 required positional argument: 'self'

[hesreallyhim]

not sure if this is the same as what you're asking, but you can specify certain parameters should be "Injected" and therefore hidden/ignored by the LLM when it's looking at the tool function signature. Check out here: https://python.langchain.com/docs/how_to/tool_runtime/#hiding-arguments-from-the-model
So maybe you can just "hide" the cls/self argument(?)

[hesreallyhim]

hmm, also maybe check this out? #9404 (comment)

[gilbertorradecki]

I'm using the same code as above, except for the model: I'm using llama3-groq-tool-use 8B locally.
I'm wodering why I couldn't achieve the same result: when I pass two requests to my model, it only prints one of them:

query = "What is 3 * 12? Also and what is 11 + 49?"

chain = llm | PydanticToolsParser(tools=[multiply, add])

chain.invoke(query)

chain.invoke(query)

Out[81]: [multiply(a=3, b=12)]