The MJX tutorial has a partial example, but it's incomplete, as it relies on a make_inference_fn that was created during training and never saved. We should have some utilities to make saving and loading easy, particularly for non-default policy architectures.
The MJX tutorial has a partial example, but it's incomplete, as it relies on a
make_inference_fn
that was created during training and never saved. We should have some utilities to make saving and loading easy, particularly for non-default policy architectures.