Lack of Guidance on Optimizing/Finetuning ReAct Agent with Few-shot Examples #703
Description
The current ReAct documentation lacks clear instructions on optimizing or finetuning a ReAct agent using few-shot examples. Both the main ReAct documentation ReAct Docs and the examples documentation Examples Docs do not provide sufficient guidance in this regard. It's essential to understand that for the ReAct agent to effectively learn from few-shot examples, the complete ReAct cycle (Question, Action, Action Input, Observation) should be encapsulated within these examples.
The provided example in the documentation, such as:
qa_pair = dspy.Example(question="This is a question?", answer="This is an answer.")
does not demonstrate the correct way to optimize or finetune a ReAct agent with few-shot examples.
Could someone please provide a clear example demonstrating the correct approach to optimizing or finetuning a ReAct agent, particularly with few-shot examples? This would greatly benefit users seeking to leverage ReAct effectively.