user avatar
Arthur Zucker
@art_zucker
Head of transformers @huggingface 🤗
Joined October 2021
649
Following
8,225
Followers
  • Pinned
    user avatar
    Arthur Zucker
    @art_zucker
    May 15, 2025
    A quick update on the future of the `transformers` library! In order to provide a source of truth for all models, we are working with the rest of the ecosystem to make the modeling code the standard. A joint effort with vLLM, LlamaCPP, SGLang, Mlx, Qwen, Glm, Unsloth, Axoloth,
    175K
  • user avatar
    Arthur Zucker
    @art_zucker
    Jul 18, 2022
    Hey! 🎶🤗@OpenAI's jukebox model will soon be released on the @huggingface hub! Before that I would like to generate a few songs! Feel free to suggest an artist, a genre and a few lyrics and I will try to see what I can do! 🤗🎶
  • user avatar
    Arthur Zucker
    @art_zucker
    Oct 11, 2022
    ⏰➡👆We just released Whisper in 🤗 transformers! @openai’s latest speech recognition transformer trained on 680 000 hours of audio! For example use case, check this notebook : colab.research.google.com/drive/16HO7if9…
  • user avatar
    Arthur Zucker
    @art_zucker
    Jan 25, 2023
    Super glad to announce that we finished refactoring 🤗's Whisper to match the API from @OpenAI ! You can now easily set the languages and task, while benefitting from 6x faster inference! 🚀🚀🚀 Feel free to ask questions!
    160K
  • user avatar
    Arthur Zucker
    @art_zucker
    Apr 19, 2022
    🤗🎶As of today I am joining @huggingface as a Machine Learning Research Engineer!! I will be working on #deeplearningformusic ! 🎶 🤗 Feel free to send me your favorite AI generated songs! 🧑🏻‍💻
  • user avatar
    Arthur Zucker
    @art_zucker
    Jan 22, 2025
    From the bottom of my heart, thanks to @deepseek_ai: this week, open source has won. We are finally showing closed AI how it’s done! 🚀
    24K
  • user avatar
    Arthur Zucker
    @art_zucker
    Nov 7, 2023
    Hey @elonmusk, anyway I can help get the Grok on 🤗?
    112K
  • user avatar
    Arthur Zucker
    @art_zucker
    Mar 3, 2023
    Wow! Google released another GREAT model on the @huggingface hub 🤗 ! Thanks a lot @googlebrain, @YiTayML and the rest of the team! How to use it? 🍮
    60K
  • user avatar
    Arthur Zucker
    @art_zucker
    Mar 19, 2025
    Small util merged in transformers, open to contribution to extend it to all models! For now I tested `gemma3`, `gemma2`, `paligemma` and `mistral`! Curious to see some of the more special ones 👀 (mllama? Qwen-Audio? Whisper? Qwen-VL?)
    36K
  • user avatar
    Arthur Zucker
    @art_zucker
    Apr 24, 2023
    CLAP is to Audio what CLIP is to Images 🎙️ Here's how easy it is to run zero-shot audio classification in transformers 🤗 :
    73K
  • user avatar
    Arthur Zucker
    @art_zucker
    Feb 16, 2024
    🤗Transformers welcomes 4x speed-up by leveraging torch.compile to its very core ! Static Cache + compile for decoder models: on llama 7b ⤵️
    00:00
    46K
  • user avatar
    Arthur Zucker
    @art_zucker
    Jan 10, 2025
    Transformer v4.48 is out! Pretty big because we added a refactor to make sure all new models can be used out of the box in vllm! And a lot of already existing models as well. Can't wait to have the ~300 architecture fully done!
    36K
  • user avatar
    Arthur Zucker
    @art_zucker
    Apr 5, 2025
    TLDR of LLama4: - 10m context length. -? memory requirements? no problem they use chunked attention (block mask) on layers that do apply rope (3 in 4). This means only 8K tokens are kept in cache for most of the layers!!!!! - mega smart irope, so qk norm post rope, and query
    45K
  • user avatar
    Arthur Zucker
    @art_zucker
    Jun 24, 2022
    Took a while but the Flax and Tensorflow weights are now available for all OPT models (until OPT-30B!) on the main branch of transformers @huggingface ! 😃