The community member is looking for the parser/code that is used to fetch function arguments when using models that don't have function calling, specifically through a ReAct agent. Another community member responds that a ReAct agent is prompted with the current tools/descriptions and chat history, and the language model has to respond with a structure that the framework can parse. They provide a link to a relevant code file. The second community member mentions that they were looking for the parse_action_reasoning_step function and its associated functions in the output_parser.py file, but had missed it previously.
Can anyone point to the parser/code that is currently being used to fetch function arguments when using models that don't have function calling (through a ReAct agent in particular)?
But a react agent is just prompted with the current tools/descriptions and chat history, and the llm has to respond with a structure that the framework can parse
I was looking for the parse_action_reasoning_step function and it's associated functions in the output_parser.py file, missed it when I was looking previously 🤷
