optax 0.2.7

A gradient processing and optimization library in JAX.

Stars: 2185, Watchers: 2185, Forks: 313, Open Issues: 65

The google-deepmind/optax repo was created 5 years ago and the last code push was 1 weeks ago.
The project is very popular with an impressive 2185 github stars!

How to Install optax

You can install optax using pip

pip install optax

or add it to a project with poetry

poetry add optax

Package Details

Author

None

License

None

Homepage

None

PyPi:

https://pypi.org/project/optax/

GitHub Repo:

https://github.com/deepmind/optax

Classifiers

• Scientific/Engineering/Artificial Intelligence
• Software Development/Libraries/Python Modules

Related Packages

Errors

A list of common optax errors.

Code Examples

Here are some optax code examples and snippets.

GitHub Issues

The optax package has 65 open issues on GitHub

• [Feature Request] Add MADGRAD Optimizer
• Add madgrad optimizer
• Implement SOAP optimizer
• Upgrade GitHub Actions to latest versions
• clipping: support default unitwise_norm for 5D params
• Implementing AdaMuon
• Add example demonstrating the microbatching api and comparing it against optax.MultiSteps.
• Upgrade GitHub Actions for Node 24 compatibility
• Remove TF dependency in Lookahead Optimizer on MNIST example
• Gradient Accumulation Results
• implements Mars optimizer [contrib]
• [Feature Request] Adding the MARS Optimizer (Variance Reduction) from Hu et al. 2024
• Raise error on unused extra kwargs in backtracking linesearch
• Add regression test for scale_by_rms zero-gradient stability
• Add semi-supervised losses and test

See more issues on GitHub