Automate Dataset Creation for Llama-2 with GPT-4!

แชร์
ฝัง
  • เผยแพร่เมื่อ 19 ต.ค. 2024

ความคิดเห็น • 30

  • @engineerprompt
    @engineerprompt  ปีที่แล้ว

    Want to connect?
    💼Consulting: calendly.com/engineerprompt/consulting-call
    🦾 Discord: discord.com/invite/t4eYQRUcXB
    ☕ Buy me a Coffee: ko-fi.com/promptengineering
    |🔴 Join Patreon: Patreon.com/PromptEngineering

  • @andyjm2k
    @andyjm2k ปีที่แล้ว +4

    I was under the impression that the openai terms of service prevents you from using GPT4 to train another model.

    • @Timotheeee1
      @Timotheeee1 ปีที่แล้ว

      only if your new model competes with them

  • @sahulsumra7025
    @sahulsumra7025 ปีที่แล้ว +1

    It's really good content

  • @kalhosni
    @kalhosni ปีที่แล้ว

    Thanks a lot. However, I got "NameError: name 'load_dataset' is not defined: at the top of "Load Datasets and Train" section. The path is correct and your copy seemed to work just fine???

  • @alx8439
    @alx8439 ปีที่แล้ว +3

    Also without the comparison to the original non-fined-tuned Llama, the effectiveness of this fine tuning says nothing. It's like adding a new mechanical part to your car and measure max speed without making the same measurement before the installation

  • @alx8439
    @alx8439 ปีที่แล้ว +1

    What was the purpose of trying to use NousResearch model instead of the original Llama model by Meta?

    • @UncleDao
      @UncleDao ปีที่แล้ว +1

      The two models are similar, except that if using Llama model, you need to login hugging face to download, if using NousResearch model, there is no need.

  • @WeylandLabs
    @WeylandLabs ปีที่แล้ว +1

    Very Nice !

  • @ludwigvanbeethoven61
    @ludwigvanbeethoven61 ปีที่แล้ว

    nice, is it also possible to run a LLM on Visual Studio Code instead of Jupiter?

    • @engineerprompt
      @engineerprompt  ปีที่แล้ว +1

      Yes, the process will be similar

  • @alejandrofernandez3478
    @alejandrofernandez3478 ปีที่แล้ว

    I wonder if soon an llm could be taught to train itself on whatever it wanted or was told to learn

    • @texasfossilguy
      @texasfossilguy ปีที่แล้ว

      youd have to have a lot of money or a very powerful architecture

    • @texasfossilguy
      @texasfossilguy ปีที่แล้ว

      it also is important to note, it doesn't want, feel, or need anything unless you somehow program that into it.

  • @lrkx_
    @lrkx_ ปีที่แล้ว

    Would you be able to fine-tune a 7B parameter model locally on a MacBook Pro M2 with enough RAM? Really curious to see if it’s possible…

    • @engineerprompt
      @engineerprompt  ปีที่แล้ว +1

      I haven’t had luck with it, so probably not

    • @satyamgupta2182
      @satyamgupta2182 ปีที่แล้ว

      @@engineerprompt did you come across a solution?

  • @troywoodson7355
    @troywoodson7355 ปีที่แล้ว +1

    NousResearch/llama-2-7b-chat-hf is no longer available. What model would you recommend instead?

    • @engineerprompt
      @engineerprompt  ปีที่แล้ว +1

      If you have the access token then use the meta original version

  • @YassineMeche
    @YassineMeche ปีที่แล้ว +1

    thank you for all this great job , and where I can find discord group

    • @engineerprompt
      @engineerprompt  ปีที่แล้ว +1

      In the description of the video :)

  • @RidvanGER
    @RidvanGER ปีที่แล้ว

    Is it possible to use any kind of source code(python, c++ or c#) as a dataset to train llama 2 or any other LLM? Thank you!

    • @engineerprompt
      @engineerprompt  ปีที่แล้ว +1

      Yes, but you will need a relatively large dataset for it to train properly.

  • @mastamindchaan387
    @mastamindchaan387 ปีที่แล้ว

    Nothing to do with a "Dataset".
    This is "how to train ChatGPT to respond to a specific prompt"

  • @azzorkimaroc4879
    @azzorkimaroc4879 ปีที่แล้ว +1

    first comment

  • @tareksahalia
    @tareksahalia ปีที่แล้ว

    السنة من أعطى قيمة لهذه الطائفة