Keras

Keras is a deep learning API designed to be easy for people to use, with simple, consistent commands and clear error messages. It runs on several backends, so models can move between JAX, TensorFlow and PyTorch. It has extensive documentation and developer guides.

What Keras offers

• Model building APIs — the Functional API and subclassing for layers and models.
• Training and evaluation — built-in methods for fitting and testing models.
• KerasHub — implementations of popular model architectures with pretrained checkpoints.
• Deployment options — models can be exported to run in the browser, on mobile devices and embedded hardware, or served through a web API.
• Code examples — guides for computer vision and natural language processing.

Who it works with

Machine learning researchers and engineers, and university courses teaching deep learning. The project says organisations such as CERN, NASA and NIH use it.

Key facts

• Licence: open source
• Backends: JAX, TensorFlow, PyTorch