We code Stanford's ALPACA LLM on a Flan-T5 LLM (in PyTorch 2.1)

Поділитися
Вставка
  • Опубліковано 15 січ 2025

КОМЕНТАРІ • 23

  • @JohnLangleyAkaDigeratus
    @JohnLangleyAkaDigeratus Рік тому +3

    Thanks for doing the live coding exercise and leaving the mistakes in.
    Sometimes. I wonder if it's only me who has experiences like that😅
    Thanks again!

  • @nosxr9732
    @nosxr9732 Рік тому +1

    Why dont you give your colab code link too? :\

  • @waeldimassi3355
    @waeldimassi3355 Рік тому

    Great videos ! The new cool kid in the neighbourhood

  • @Whisper_InThe_Rain
    @Whisper_InThe_Rain Рік тому

    Do you have the Google collab notebook for this?

  • @ziad12211
    @ziad12211 Рік тому +2

    Next time i hope the video about Alpaca lora

    • @EmmanuelMadu
      @EmmanuelMadu Рік тому

      Same, it would be great to see a breakdown on how to train one

  • @issachicks1
    @issachicks1 Рік тому

    Love your videos, been following for a while.
    Have you done any benchmarking to compare the Flan-T5 based ALPACA to the LLaMa based ALPACA? Curious to know what the final performance of the Flan-T5 open model is in comparison to the original ALPACA model

    • @code4AI
      @code4AI  Рік тому +1

      Have to build the channel first, then activate monetization (guess currently its about 5cents per video per month), all additional expenses for Azure 640GB Alpaca benchmarks are second in list.

  • @web3digitalmarketingpreneur

    Your videos are a really great man. Am pretty new to these LLMs and starting to get around the idea of this AI revolution. What would be the best way to reach out to you if I had any questions in mind🤔?

  • @ml0k1
    @ml0k1 Рік тому

    great video mate. Sorry to ask but can you provide the COLAB? Thanks once again for your knowledge. Cheers

    • @code4AI
      @code4AI  Рік тому +1

      As mentioned, code sequences are available as python files to download.

  • @KevinRohling-o8h
    @KevinRohling-o8h Рік тому

    This is amazing. Huge thanks for this. Any chance you've got a link to the notebook? Btw, I'm getting a CUDA out of memory error on a machine with 24GB of GPU RAM. Any chance you've got a pointer I might be missing?

    • @KevinRohling-o8h
      @KevinRohling-o8h Рік тому

      Welp, I just noticed you're training a t5-large and I was trying a t5-xlarge so it looks like I answered my own question there :) Anywho, if you happen to have the notebook link that would be cool. Again, this is awesome.

    • @nosxr9732
      @nosxr9732 Рік тому

      can you please share your colab code, im beginner in this and finding this very difficult to start with. Thanks :)

  • @sirrr.9961
    @sirrr.9961 Рік тому

    I am huge fan of your videos. i am non programmer but have big interest in this stuff. I wanna request to actually give us a walkthrough on how to prepare our own data like some image pdf files to convert into vector embeddings and use contextual injections to make our own bots even for personal purposes. One more thing I am confused about is how to prepare examples to train our AI to do specific tasks like writing a report in a specified manner and use the specified vocabulary. Is there any written resource where I could learn that? Please reply. 😊

    • @code4AI
      @code4AI  Рік тому

      If you are not a programmer like you say then watching videos to code your own AI system has a steep learning curve. If you want to start, get familiar with the transformer architecture of AI models, every i/o question will then be answered.

  • @riser9644
    @riser9644 Рік тому

    Can we use it for text classification

  • @Larzsolice
    @Larzsolice Рік тому

    Please do this for the new Dolly Dataset. That would be epic

  • @jackbauer322
    @jackbauer322 Рік тому +1

    I didn't understand a THING ! What is this ? What is it used for ? CONTEXT PLEASE !

    • @code4AI
      @code4AI  Рік тому +1

      Easy.
      Flan-T5 LLM is explained here: ua-cam.com/video/SHMsdAPo2Ls/v-deo.html
      Alpaca is explained here: ua-cam.com/video/j6dqO2dSF9c/v-deo.html
      Enjoy.

    • @p-j-y-d
      @p-j-y-d Рік тому

      The channel IS the context. Not all series have to do the "Previously on 24" thing, Jack Bauer.

  • @p-j-y-d
    @p-j-y-d Рік тому

    How much $$$ did the training cost you?

    • @code4AI
      @code4AI  Рік тому

      If you run the small PLM on a free COLAB T4, nothing. But for bigger LLMs I have a new video upcoming with 8bit quantization and PEFT-LoRA which would allow for a faster and cheaper fine-tuning.