* feat(train): FSDP checkpoint saving
* adding docs for FSDP
* adding a test for the fsdp checkpoint path
* cleanup
* fixing final upload to hub
* refactored initial implementation to use torch fsdp api and adding new tests
- Guard ACT's KL divergence computation against None latent params to
prevent crashes during eval when use_vae is set but the forward path
returns no VAE outputs.
- Add offline batch fallback to Diffusion's predict_action_chunk() so
it works with dataloader batches (empty queues) in addition to the
existing online rollout path (populated queues). This enables batched
action prediction for offline evaluation.
- Add pretrained_revision field to PreTrainedConfig (policies) and
RewardModelConfig (reward models), and thread it through make_policy(),
make_pre_post_processors(), and make_reward_model() so that weights and
processor configs can be loaded from a specific Hub commit, branch, or
tag. Defaults to None (latest version, preserving current behavior).
Dataset and env hub loading already supported revision pinning.
Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>
* fix(logging): batch wandb metrics
- Batch all metrics into a single wandb.log() call instead of one per
key, reducing API overhead.
- Add support for list-valued metrics by expanding them to indexed keys (e.g.
metric_0, metric_1).
* fix(stats): handle scalar stats robustly
- Wrap cast_stats_to_numpy with np.atleast_1d to prevent 0-d arrays
from scalar stats causing shape mismatches downstream.
* fix(logging): remove unused list-valued metric expansion
---------
Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>
* feat(envs): add env plugin discovery
- Add 'lerobot_env_' to third-party plugin discovery prefixes, completing
the plugin system for all component types (robots, cameras, teleoperators,
policies, and now environments). External packages named lerobot_env_*
can self-register EnvConfig subclasses on import, enabling --env.type=
resolution without lerobot code changes.
* feat(envs): add generic observation passthrough
- Add generic observation passthrough in preprocess_observation() for
unhandled ndarray/tensor keys, replacing the pattern of adding per-env
hardcoded key handlers. Extra keys are forwarded as observation.<key>
and can be shaped by env-specific ProcessorSteps via get_env_processors().
---------
Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>
CarolinePascal