JAX

High-performance numerical computing. Composable transformations for autograd, JIT, and vectorization.

About JAX

JAX is a Python library by Google for high-performance numerical computing and machine learning research. It provides composable transformations of Python+NumPy programs: differentiate, vectorize, parallelize, and JIT-compile to GPU/TPU.

Key Features

Autograd on native Python/NumPy
XLA JIT compilation
vmap for automatic vectorization
pmap for multi-device parallelism
Functional programming paradigm
TPU & GPU native support

Why choose JAX?

JAX is an open source alternative to PyTorch, TensorFlow. Licensed under Apache-2.0, it gives you full access to the source code and the freedom to modify, self-host, and contribute. It is available as a desktop or web application.