AI BriefingHugging Face and KerasHub Integration: Transformers Model Support
·2024.07.10 09:00
Key point
Transformers models from Hugging Face can now be loaded and used directly in KerasHub.
Details
Over 300,000 models based on the Transformers library stored on the Hugging Face Hub can now be loaded and used directly in KerasHub.
Previously, KerasHub users were limited to using Keras-only models, but with this integration, major models such as Gemma (1, 2), Llama 3, PaliGemma can now be used immediately.
Key features and benefits are as follows.
- Framework flexibility: Through KerasHub's backend support, Transformers models can be run immediately in TensorFlow, JAX, PyTorch—whichever environment you prefer—or converted to TFLite and other formats for deployment.
- Automatic conversion: Configuration files, weights (safetensors), and tokenizers are internally mapped, allowing users to use models without a separate conversion process.
- Usage: By updating the
keras-hubandkeraslibraries to the latest versions, models can be loaded simply via thefrom_presetmethod.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.