Get started with HuggingFace Transformers - Pipeline, Custom Pipeline, Tokenizer, Model, Hub

Поділитися
Вставка
  • Опубліковано 3 сер 2024
  • The Transformers architecture is fundamental to almost all the Large Language Models (LLMs) out there like Chat-GPT, LLAMA, Vicuna and you name it. The transformers is the main building block of these models.
    If you want to get hands-on and train or fine-tune LLMs you cannot escape working with the Transformers Python library from HuggingFace.
    So In this video, I will walk you through the Transformers library so that you can get up to speed working with it.
    ⌚️ ⌚️ ⌚️ TIMESTAMPS ⌚️ ⌚️ ⌚️
    0:00 - Intro
    1:05 - Installation
    1:25 - Pipeline
    3:46 - Create Custom Pipeline
    5:59 - AutoClass
    8:28 - Tokenizer
    11:30 - Model
    13:39 - Saving the Model
    14:55 - HuggingFace Hub
    15:22 - Extro
    RELATED LINKS
    Notebook for this video: github.com/ai-bites/generativ...
    HuggingFace Transformers: huggingface.co/docs/transform...
    Transformers github repo: github.com/huggingface/transf...
    MY KEY LINKS
    UA-cam: / @aibites
    Twitter: / ai_bites​
    Patreon: / ai_bites​
    Github: github.com/ai-bites​
    #machinelearning #deeplearning #aibites

КОМЕНТАРІ • 8

  • @frazuppi4897
    @frazuppi4897 5 місяців тому +1

    hugging face transformers = spaghetti code :)

    • @AIBites
      @AIBites  5 місяців тому +1

      hah.. thanks for your comment. like it or not, most people are using it. Would you recommend any other library to say fine-tune an LLM?

    • @frazuppi4897
      @frazuppi4897 5 місяців тому

      @@AIBites maybe lighting stuff idk I worked on hf and man the code is radioactive - at some point some else will take over and make a better framework. But the user facing APIs are great, cannot say I'd want to trave place with the people working there now

    • @AIBites
      @AIBites  5 місяців тому +1

      I have worked with Lighting on the other hand and it's a debugging nightmare. For me, implementing everything in PyTorch wins over anything else!

    • @frazuppi4897
      @frazuppi4897 5 місяців тому

      @@AIBites oh really? I've been using lighting for a long time, what was your issue? Curious to hear it, I also have a lot of contacts there so I'd be happy to share it with them and hopefully they can come up with something better

    • @AIBites
      @AIBites  4 місяці тому +1

      looks like the community is moving fast towards HF now. For example, Google releasing its gemma models on HF hub and also collaborating with HF.
      I am just making videos on whichever could get more viewership :)