JSONCheckpointAdapter#

class torchrl.checkpoint.JSONCheckpointAdapter[source]#

Adapter for JSON-compatible configuration, metrics, and metadata.

Mutable mappings and lists are updated in place during restoration. Other values are returned through CheckpointLoadResult.values.

Examples

>>> from torchrl.checkpoint import JSONCheckpointAdapter
>>> JSONCheckpointAdapter().adapter_id
'torchrl.json'

load(component: Any, path: Path, *, map_location: Any, tensor_load_kwargs: Mapping[str, Any], args: tuple[Any, ...], kwargs: Mapping[str, Any]) → Any[source]#: Restore component from path and optionally return a value.

save(component: Any, path: Path, *, args: tuple[Any, ...], kwargs: Mapping[str, Any]) → Any[source]#: Save component below path.

Docs