Train your JAX models using model.fit(...) in Keras 3

Опубликовано 22 мая 2025, 3:22

The Keras ecosystem offers many on-ramps to JAX. Use our classic and well-loved API patterns such as `model.fit(...)` to author and train models using JAX. Want to start with a pre-trained model? Try KerasHub, our catalog of 100+ model architectures with support for thousands of HuggingFace checkpoints. Fine-tune and optimize them using LoRA and qLoRA with JAX (or your choice of backend). We’ll also introduce Keras Recommenders, the latest addition to the Keras ecosystem.

Resources:
Keras → goo.gle/4d9t3LL
Keras Hub → goo.gle/4iYQEQp
Keras Recommenders → goo.gle/4mr8YoD
Keras on Kaggle → goo.gle/44P06CT
Keras on Hugging Face → goo.gle/42RLuBd
Keras developer guides → goo.gle/437YNMJ

Speakers: Divyashree Sreepathihalli , Monica Song, Fabien Hertschuh

Check out the AI session track from Google I/O 2025 → goo.gle/io25-ai-yt
Check out all of the sessions from Google I/O 2025→ goo.gle/io25-sessions-yt

Subscribe to Google for Developers → goo.gle/developers

Event: Google I/O 2025

Products Mentioned: AI/Machine Learning