Hugging Face Hub and KerasHub integration enables loading Transformers models in KerasHub (Gemma, Llama3, PaliGemma)

AI Impact Summary

Hugging Face Hub and KerasHub now share a common model save format, allowing Transformer checkpoints to be loaded directly into KerasHub. The initial rollout supports Gemma (1 and 2), Llama3, and PaliGemma, with 33 models in this first batch. Transformers checkpoints can be loaded into KerasHub and deployed across TensorFlow, JAX, or PyTorch backends with the same workflow. This expands the KerasHub model surface and should accelerate experimentation and time-to-value for teams relying on Hugging Face checkpoints; ensure downstream pipelines are updated to use keras-hub and keras>=3.3.3 and validate with presets like Hermes-2-Pro-Llama-3-8B.