inference

Inference backends for foundation-model forecasters.

An InferenceBackend isolates how a checkpoint is executed behind a single named-tensor contract; forecasters compose a backend rather than inheriting execution behaviour. Only the dependency-free surface (the protocol and the execution-provider configs) is re-exported here; a concrete backend lives in its own submodule and imports its heavy runtime at module top level.

Submodules

openstef_foundation_models.inference.backend	The InferenceBackend contract shared by all execution backends.
openstef_foundation_models.inference.onnx_backend	ONNX Runtime execution backend.
openstef_foundation_models.inference.provider_selection	Metadata-driven execution-provider selection.
openstef_foundation_models.inference.providers	Typed ONNX Runtime execution-provider configuration.

Classes

CoreMLProvider(**data)	The CoreML execution provider (Apple, GPU/Neural Engine).
CpuProvider(**data)	The default CPU execution provider.
CudaProvider(**data)	The CUDA (NVIDIA GPU) execution provider.
InferenceBackend(*args, **kwargs)	A model-agnostic execution backend.
SessionOptionsConfig(**data)	A subset of ONNX Runtime SessionOptions exposed as typed config.
TensorRTProvider(**data)	The TensorRT execution provider (NVIDIA, ahead-of-time engine build).