Keras Community Newsletter - April 2026

[Keras Team]

Keras Community Updates

April 2026

Welcome to the Newsletter

We're delighted to have you in the Keras community! We'll be publishing these newsletters following community meetings as a way to share updates more broadly. We hope you find it informative and interesting!

Since the last update, the team has been hard at work adding Gemma 4 to KerasHub, tons of new features in the latest 3.14 release, Orbax Checkpointing integration, and even a new library 👀, as well as making updates to keras.io.

Happy machine learning,
The Keras team

Gemma 4

Gemma 4 is now available in KerasHub!

Try it today: https://www.kaggle.com/models/keras/gemma4

• Advanced reasoning

• Agentic workflows

• Code generation

• Longer context (128K & 256K)

• Vision and audio(audio is coming soon)

• 140+ languages

Gemma 4 on Kaggle ↗︎

New KerasHub Models

Gemma Family

• Gemma 4 

• T5Gemma 2 

• Embedding Gemma 

Kaggle ↗︎

Others

• Meta CLIP2 

• VideoPrism

Kaggle ↗︎

New 3.14 release

Features

• Orbax Checkpoint Integration

• Quantization Upgrades

  • Activation-aware Weight Quantization (AWQ)

  • Asymmetric INT4 Sub-Channel Quantization

• Batch renormalization in BatchRenormalization layer

• ScheduleFreeAdamW optimizer (thanks timon0305 !)

• Gated Attention in MultiHeadAttention & GroupedQueryAttention (thanks @pass-lin !)

• CLAHE (Contrast Limited Adaptive Histogram Equalization) preprocessing layer (thanks @AlanPonnachan !)

• Many new ops (NaN-aware Numpy ops, math & linalg ops)

• Grain support in preprocessing layers (thanks @rstar327 !)

• OpenVino almost feature complete (thanks @goyaladitya05  !)

Release Notes:https://github.com/keras-team/keras/releases

Download today ↗︎

Orbax Checkpointing Integration

Keras now supports high-performance checkpointing

Take advantage of state-of-the-art checkpointing techniques that are used by Google Deepmind with Orbax Checkpointing.

Learn more ↗︎

Sneakpeak 🤫

New library: Kinetic

We announced at our community meeting on April 3 an exciting new library: Kinetic!

Expect more content and tutorials in the future!

Check it out on Github ↗︎

Keras.io Website updates

New Guides

Native Function calling guide With FunctionGemma 

• Function Calling With FunctionGemma

Object Detection Fine-tuning with KerasHub

• Object Detection with KerasHub

Tutorials Migrated to Keras 3 :

• Focal Modulation: A replacement for Self-Attention     

• Masked image modeling with Autoencoders 

• Automatic Speech Recognition using CTC 

• Distilling Vision Transformers

• Graph attention network (GAT) for node classification

• Graph representation learning with node2vec

Call for Contributions

Looking for a way to contribute to Keras? Start here!

• https://github.com/keras-team/keras-hub/issues/1836

• https://github.com/keras-team/keras/issues/19519

• https://github.com/keras-team/keras-rs/issues 

We will also be reaching out to contributors over GitHub (via email or LinkedIn)! Thank you to Alan Ponnachan for presenting at the meeting on April 3.

Keras Github

You're receiving this email because you're subscribed to the Keras Users group. To stop receiving these emails, update your subscription settings in the group page.