google-deepmind/optax
A composable gradient processing and optimization library for the JAX machine learning framework.

Velocity · 7d
+1.0
★ / day
Trend
→steady
star history
Optax provides building blocks for creating custom optimizers and gradient processing components within the JAX ecosystem. Originally developed internally at DeepMind as jax.experimental.optix, it offers implementations of popular optimizers and loss functions designed for research productivity. The library emphasizes composability, allowing researchers to combine low-level components into custom solutions.