Mastering LangChain RAG: Integrating Chat History (Part 2)

[@GopalNoutiyal]

Please continue making more video's. Appreciate your effort and contribution. Thank you

[@srirammoorthy9337]

@Eric, thank you for posting this video. Your video's are clear and easy to understand. I have been reading through the Langchain docs and did not find it friendly to understand. Please continue making more video's. Appreciate your effort and contribution. Thank you Regards, Sriram Moorthy

[@andrew.derevo]

Thanks a lot for video. It’s interesting how exactly chain pass the messages to LLM. I’m not 100% sure but I think they need extra formatting.

[@sangram7153]

Really great vedios!! Thank You so much

[@BertrandGoetzmann]

Hello Eric, thank you very much for this video. It's interesting to understand how the question can be contextualized using history to query the vectorstore; however, is there a really interest to use the history in the qa_prompt. Isn't enough to just have the context and the question? What do you think?

[@eric_vaillancourt]

The history is important when you ask a follow up question.

[@BertrandGoetzmann]

@@eric_vaillancourt I agree. I supposed the input injected in the qa_prompt came from the reformulated question. So I suppose it's not the case.

[@sangram7153]

Which approach is reliable to use? this one or using lang-chain Conversation memories like Conversation Buffer Memory Conversation Buffer Window Memory Conversation Summary Memory Conversation Summary Buffer Memory

[@eric_vaillancourt]

It all depends on your needs. In my experience, summarizing is a good technique.

[@sangram7153]

Have some quries? Lets say if we ask 100+ question so will it cause a LLM context limit??? are you passing the chat history in prompt to LLM??

[@eric_vaillancourt]

At some point, you will have to do some memory management, like summarizing the conversation.

[@lesptitsoiseaux]

Fameux!

[@sangram7153]

How many numbers of converations are passed in prompt every time?? can we change that number? how?

[@eric_vaillancourt]

You decide. It also depends on the context limit of the LLM