• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Wightman Explains OpenCLIP ModernText and ModernBERT Differences

    AI engineer answers query on native versus Transformers encoders in OpenCLIP.

    RW
    1 Source, 27d ago, first seen 27d ago

    TLDR

    Ross Wightman, creator of the timm PyTorch library, responded publicly to a question about recent OpenCLIP work. The query asked for differences between the native ModernText encoder and supported pretrained ModernBERT models available through Transformers. He stated that both are driven by similar goals. The native implementation provides additional configuration options and simpler customization, while the Transformers integration supplies solid pretrained models. The exchange appears among visible replies tied to his post.

    Combined views

    2.2K

    1 Source, first seen 27d ago

    Combined views

    2.2K

    1 Source, first seen 27d ago

    11 likes
    11 likes
    1 comments
    15 saves

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    1 comments
    15 saves

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    1 Source

    @wightmanrSomeone working with new OpenCLIP additions asked me about the differences between the native 'ModernText' encoder and the supported 🤗 Transformers (pretrained) ModernBERT models. I provided a brief answer. Both inspired by the same goals, but there are more config knobs and easier customization (native) vs solid pretrained weights but more pooling constraints (transfomers). Then I asked Fable 5.1 for a run down (you know brief), it went a bit over the top, but it's actually a useful comparison for people looking at the new features in OpenCLIP. See reply. If you've got any ideas for additional ModernText adds, it's easy to customize. It's also not just an encoder, the ModernText additions were intended to be used as encoder and/or decoder for models like MaMMUT (both).

    1 Source

    @wightmanrSomeone working with new OpenCLIP additions asked me about the differences between the native 'ModernText' encoder and the supported 🤗 Transformers (pretrained) ModernBERT models. I provided a brief answer. Both inspired by the same goals, but there are more config knobs and easier customization (native) vs solid pretrained weights but more pooling constraints (transfomers). Then I asked Fable 5.1 for a run down (you know brief), it went a bit over the top, but it's actually a useful comparison for people looking at the new features in OpenCLIP. See reply. If you've got any ideas for additional ModernText adds, it's easy to customize. It's also not just an encoder, the ModernText additions were intended to be used as encoder and/or decoder for models like MaMMUT (both).