unilab.algos.torch.appo.runtime

Runtime resolution helpers for APPO script assembly.

This module keeps entrypoint scripts generic: they resolve an APPO runtime bundle from owner config and then call the returned runner/play entrypoints without knowing which concrete runtime implementation is active.

Functions

resolve_appo_runtime(rl_cfg, *, default_play_fn)

Resolve the APPO runtime bundle from owner config.

Classes

APPORuntime

Resolved APPO runtime entrypoints consumed by the generic script.
class unilab.algos.torch.appo.runtime.APPORuntime[source]

Bases: object

Resolved APPO runtime entrypoints consumed by the generic script.

Parameters:

  • runner_cls (type[Any]) – Runner class used for APPO training mode.

  • play_fn (Callable[..., str | None]) – Play-mode callable used for checkpoint playback.

Returns:

Immutable APPO runtime bundle selected from owner config.

runner_cls: type[Any]
play_fn: Callable[..., str | None]
__init__(runner_cls, play_fn)
Parameters:
unilab.algos.torch.appo.runtime.resolve_appo_runtime(rl_cfg, *, default_play_fn)[source]

Resolve the APPO runtime bundle from owner config.

Parameters:

  • rl_cfg (dict[str, Any]) – Resolved algorithm config dictionary from Hydra composition.

  • default_play_fn (Callable[..., str | None]) – Generic APPO play function used when no custom runtime resolver is selected by the owner config.

Return type:

APPORuntime

Returns:

APPORuntime containing the train and play entrypoints for the selected APPO runtime.