vertebrae.extractors.onnx

Optional ONNX runtime feature extractor.

Classes

ONNXExtractor

Wrap a local ONNX model as a feature extractor.

Module Contents

class vertebrae.extractors.onnx.ONNXExtractor(name, model_path, input_fn=None, output_fn=None, outputs=None, structured_outputs=None, input_names=None, output_names=None, providers=None, provider_options=None, modality='unknown', extractor_type='custom_onnx', recipe_data=None, allow_sparse=False, streaming_safe=True, external_data_paths=None, cache_identity=None)[source]

Wrap a local ONNX model as a feature extractor.

Parameters:

  • name (str) – User-facing extractor name.

  • model_path (Union[str, pathlib.Path]) – Path to a local ONNX model file.

  • input_fn (Optional[Callable[[Any], Any]]) – Optional callable that converts raw inputs into ONNX inputs.

  • output_fn (Optional[Callable[[Sequence[Any]], Any]]) – Optional callable that converts raw ONNX outputs into embeddings.

  • input_names (Optional[List[str]]) – Optional ONNX input names to feed. Defaults to the model inputs.

  • output_names (Optional[List[str]]) – Optional ONNX output names to fetch. Defaults to the model outputs.

  • providers (Optional[List[str]]) – Optional ONNX Runtime execution providers.

  • provider_options (Optional[List[Dict[str, Any]]]) – Optional provider-specific configuration dictionaries.

  • modality (str) – Input modality metadata.

  • extractor_type (str) – Extractor family metadata.

  • recipe_data (Optional[Dict[str, Any]]) – Extra serializable metadata for reproducibility.

  • allow_sparse (bool) – Whether sparse embedding outputs are allowed.

  • streaming_safe (bool) – Whether independent batches can be embedded without full-context state.

  • outputs (Optional[List[Dict[str, Any]]])

  • structured_outputs (Optional[List[Dict[str, Any]]])

  • external_data_paths (Optional[Iterable[str]])

  • cache_identity (Optional[str])

fit(X, y=None)[source]

No-op fit for frozen ONNX models.

Parameters:

  • X (Any)

  • y (Any)

Return type:

ONNXExtractor

transform(X)[source]

Run the ONNX model and validate the resulting embeddings.

Parameters:

X (Any)

Return type:

numpy.ndarray

fit_transform(X, y=None)[source]

Run the ONNX model and return validated embeddings.

Parameters:

  • X (Any)

  • y (Any)

Return type:

numpy.ndarray

recipe()[source]

Return a serializable ONNX extractor recipe.

Return type:

Dict[str, Any]

get_resource_profile_adapter()[source]

Return ONNX Runtime model-artifact profiling hooks.

Return type:

Any