espnet3.parallel.inference_provider.InferenceProvider

source

class espnet3.parallel.inference_provider.InferenceProvider(config: DictConfig, , params: Dict[str, Any] | None = None)

Bases: EnvironmentProvider, ABC

Provider base class tailored for inference-time datasets/models.

Subclasses implement dataset/model creation for local execution or per-worker setup. Instances cache a local environment for reuse.

Initialize the provider and prebuild the local environment.

• Parameters:
  • config – Configuration for dataset/model construction.
  • params – Optional extra entries merged into the environment.

abstractmethod static build_dataset(cfg: DictConfig)

Create a dataset object from config.

• Parameters:cfg – Configuration for dataset construction.
• Returns: Dataset-like object used for inference.

build_env_local() → Dict[str, Any]

Return the already constructed local environment.

• Returns: Mapping with dataset/model and any extra params.

abstractmethod static build_model(cfg: DictConfig)

Create a model object from config.

• Parameters:cfg – Configuration for model construction.
• Returns: Model-like object used for inference.

make_worker_setup_fn()

Return a setup function that rebuilds the env per worker.

• Returns: Callable that constructs a fresh environment when invoked.