You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We are currently using the brax.training.ppo training scripts in our project and have been very happy with Brax's performance.
We noticed that the brax.training module uses the older flax.linen API for neural network definitions. Flax recently introduced the flax.nnx API, which offers a more Pythonic and streamlined approach to model development.
Are there any plans to transition the brax repository to use the nnx API in the future? This would help improve model flexibility and maintainability for projects that depend on brax.training.
Thank you for your time and consideration.
Scott
The text was updated successfully, but these errors were encountered:
Hi @scott-yj-yang there aren't immediate plans to switch to nnx as flax.linen isn't being deprecated, but we'd be happy to review clean/minimal PRs. nnx does generally look like a cleaner API.
Hi Brax development team,
We are currently using the
brax.training.ppo
training scripts in our project and have been very happy with Brax's performance.We noticed that the
brax.training
module uses the olderflax.linen
API for neural network definitions. Flax recently introduced theflax.nnx
API, which offers a more Pythonic and streamlined approach to model development.Are there any plans to transition the brax repository to use the
nnx
API in the future? This would help improve model flexibility and maintainability for projects that depend onbrax.training
.Thank you for your time and consideration.
Scott
The text was updated successfully, but these errors were encountered: