keras: a multi-backend deep learning framework that eliminates framework lock-in by supporting JAX, TensorFlow, and PyTorch

What it solves

Keras 3 is a high-level deep learning framework designed to eliminate framework lock-in. It allows developers to build and train models for various tasks—such as computer vision, NLP, and audio processing—while remaining agnostic to the underlying engine that actually executes the computations.

How it works

Keras 3 acts as a multi-backend interface. It provides a consistent high-level API for model development, which can then be run on top of JAX, TensorFlow, PyTorch, or OpenVINO (the latter being for inference only). Users can switch between these backends by configuring an environment variable, allowing them to leverage the specific performance or ecosystem advantages of each framework without rewriting their code.

Who it’s for

It is designed for deep learning practitioners, from individual developers to global enterprises, who want to accelerate model development and avoid being tied to a single backend framework.

Highlights

• Multi-backend support: Compatible with JAX, TensorFlow, and PyTorch for training, and OpenVINO for inference.
• Performance optimization: Ability to pick the fastest backend for a specific architecture, potentially increasing speed by 20% to 350%.
• Interoperability: Keras models can use datasets from any format, including tf.data.Dataset and PyTorch DataLoaders.
• ** wysokie scalability**: Scales from local laptops to large GPU or TPU clusters.