Conversational Memory for LLMs Using LangChain and Huggingface - Python

Поделиться
HTML-код
  • Опубликовано: 15 ноя 2024

Комментарии • 10

  • @Ece-kx6qk
    @Ece-kx6qk 7 месяцев назад +3

    Thank you so much, a video that difficult to find on the internet again

  • @nbvcxz098
    @nbvcxz098 7 месяцев назад +1

    Thanks a lot man! This helps a lot

  • @AIEasySolutions
    @AIEasySolutions 14 дней назад

    Dear Uygar Kurt. Thank you for your videos! They are very nice. I am from Baku, Azerbaijan, Selamlar! I have a problem, maybe you know a solution. I need to create a chatbot for educational purposes, with memory, for just 10 questions to the chatbot. I have a data generated, which are dialogues, but not just one question and one answer, it is a dialogue. How can I use it for fine tuning? Can you give a guidelines. Also, I have a set of rules for the system, like prices of the products, and model also should output final cost using multiplication.

    • @uygarkurtai
      @uygarkurtai  14 дней назад +1

      @@AIEasySolutions Thank you! I can think of several approaches. You can split the dialogue and feed gradually. Or you can just feed it at once too. Model should learn. For rules maybe langchain has something.

    • @AIEasySolutions
      @AIEasySolutions 13 дней назад

      Thank you for your reply, really appreciate! Do you think feeding gradually is a good choice? Since it is a dialogue and at the end it should summarize what has been said. Anyway, thank you very much!

    • @uygarkurtai
      @uygarkurtai  13 дней назад +1

      @@AIEasySolutions Yes, while feeding gradually you can give the dialogue history as an input as well. Or if you want to use less memory you can give a summarization of it. With this I expect it to learn. Also huggingface does this very well at the try model sections next to the model. I haven't checked the way they used but you can check that out also.

    • @AIEasySolutions
      @AIEasySolutions 12 дней назад

      @@uygarkurtai Thank you, Dear Uygar! I will share my solution if I can work it out.