A small tool for making neural net training loop with flax.linen.
This library is designed to be a set of small tools and not to be a full-blown framework for deep learning. Users should keep freedom not to use some parts of this library and the rest of the library should still be able to serve the users.
Currently, this library contains the following components:
cron
: Tools for defining and performing periodical actions in the training-loop.evaluation
: Helpers for performing evaluation and handling metrics.pmap_utils
: Tools for helping the use ofjax.pmap
.pytypes
: Type definitions for better type annotations.tensorboard
: Helper functions for publishing training state and evaluation summaries.training
: Helpers for writing training-loops for Flax models.var_utils
: Utility for accessing variable collections of Flax models.
See CONTRIBUTING.md
for details.
Apache 2.0; see LICENSE
for details.
This library is still in a very early phase of development, and therefore, its API can often be subject to destructive change.
This project is not an official Google project. It is not supported by Google and Google specifically disclaims all warranties as to its quality, merchantability, or fitness for a particular purpose.