Brax 0.14.0 and updates from here

[btaba]

Hello users!

We just released brax 0.14.0, which has many key upgrades to brax PPO (see the release notes) such as fixes to adaptive KL learning rate and modifications to the running stats to work for long training runs (20B+ steps).

We actively use and continue to maintain brax/training, but other parts of brax are getting less love and we are no longer maintaining them, such as brax/io, brax/envs, the training notebooks, and the different physics backends. We may re-release brax as a pure RL library at some point in the future, depending on the usage. Thanks for all the love and support up until now!

Here are great alternatives:

• MuJoCo Playground contains a simpler env authoring framework that uses MJX directly and works great with brax RL algos!
• MJX is the canonical way to get MuJoCo simulation through JAX, including the new shiny MuJoCo Warp. See the post here about how to use MuJoCo Warp through MJX.
• Since brax joined MuJoCo, we are continuing to think about the future of GPU simulation, RL, and robotics through the MuJoCo ecosystem. Stay tuned!

❤️ the brax devs