问题

I'm utilizing OpenAI's ChatGPT model within the LangChain framework, where I pass input_documents to provide context for generating responses. However, I would like to know if there is a way to extract the specific list of input documents that were utilized by ChatGPT during the response generation process in LangChain. Either by a specific prompt to ChatGPT or using some attribute in load_qa_chain I'm not aware of. Meaning, if I passed 4 input documents and ChatGPT only used 2 of them to generate a result - which 2 did ChatGPT use?

Thank you for your assistance and any insights you can provide!

答案1

得分: 1

你是否查看了链条 RetrievalQAWithSourcesChain（https://python.langchain.com/en/latest/modules/chains/index_examples/vector_db_qa_with_sources.html），或者 Question Answering with Sources（https://python.langchain.com/en/latest/modules/chains/index_examples/qa_with_sources.html），或者通过使用 load_qa_with_sources_chain（https://python.langchain.com/en/latest/use_cases/question_answering.html）？

如果你想获得更多反馈，还可以添加 StdOutCallbackHandler（https://python.langchain.com/en/latest/modules/callbacks/getting_started.html）

handler = StdOutCallbackHandler()
chain.run(input_documents=docs, question=query, callbacks=[handler])

或者设置 verbose=True（https://python.langchain.com/en/latest/modules/memory/getting_started.html）

handler = StdOutCallbackHandler()
chain.run((input_documents=docs, question=query, callbacks=[handler])