ONNX graph ops utility¶

Collections of utilty to manipulate onnx graph.

vortex.runtime.onnx.graph_ops.embed_model_property.embed_model_property(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, props: Dict[str, Any])¶

Embed information to model, including ‘output_format’ and ‘class_names’

Parameters

model (onnx.ModelProto) – model
props (Dict[str,Any]) – information, must include ‘output_format’ and ‘class_names’

Returns

model

Return type

onnx.ModelProto

vortex.runtime.onnx.graph_ops.embed_class_names_metadata.embed_class_names_metadata(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, class_names: Union[List[str], Dict])¶

Embed class_names information to model

Parameters

model (onnx.ModelProto) – model
class_names (Union[List[str],Dict]) – class_names, either list of string or mapping from string to int

Raises

TypeError – class_names is not dict nor list

Returns

model

Return type

onnx.ModelProto

vortex.runtime.onnx.graph_ops.embed_output_format_metadata.embed_output_format_metadata(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, output_format: Dict[str, Union[List[int], int]])¶

Format each ‘output’ in ‘output_format’ and embed it to existing model

Parameters

model (onnx.ModelProto) – model
output_format (Dict[str,Union[List[int],int]]) – output_format, must contains ‘indices’ and ‘axis’ entry

Raises

TypeError – value type of output_format is not dictionary
ValueError – each entry in output_format doesnt contains ‘indices’ and/or ‘axis’

Returns

model with output_format embedded

Return type

onnx.ModelProto

vortex.runtime.onnx.graph_ops.embed_metadata.embed_metadata(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, key: str, value: Any) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Embed information to existing model

Parameters

model (onnx.ModelProto) – model
key (str) – a name representing the value
value (Any) – value to be embedded into model

Raises

TypeError – key is not string
TypeError – return value from formatter(value)

Returns

model

Return type

onnx.ModelProto

class vortex.runtime.onnx.graph_ops.embed_metadata.EmbedMetadata(key: str, value: Any)¶

__call__(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Transform model, call derived run function.

Parameters: model (onnx.ModelProto) – model to be transformed
Returns: transformed model
Return type: onnx.ModelProto

classmethod apply(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, key: str, value: Any) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Embed information to existing model

Parameters

model (onnx.ModelProto) – model
key (str) – a name representing the value
value (Any) – value to be embedded into model

Raises

TypeError – key is not string
TypeError – return value from formatter(value)

Returns

model

Return type

onnx.ModelProto

classmethod parse(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, key: str)¶

Parse metadata stored at model using json

Parameters

model (onnx.ModelProto) – model to be extracted
key (str) – metadata key

Returns

json.loads result or None if key not exists

Return type

Any

run(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Run transformation

Parameters: model (onnx.ModelProto) – model
Returns: model with embedded metadata
Return type: onnx.ModelProto

class vortex.runtime.onnx.graph_ops.embed_model_property.EmbedModelProperty(props)¶

__call__(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Transform model, call derived run function.

Parameters: model (onnx.ModelProto) – model to be transformed
Returns: transformed model
Return type: onnx.ModelProto

classmethod apply(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, props: Dict[str, Any])¶

Embed information to model, including ‘output_format’ and ‘class_names’

Parameters

model (onnx.ModelProto) – model
props (Dict[str,Any]) – information, must include ‘output_format’ and ‘class_names’

Returns

model

Return type

onnx.ModelProto

classmethod parse(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → Dict[str, Any]¶

Extract output_format and class_names from model

Parameters: model (onnx.ModelProto) – model
Returns: dictionary contains ‘output_format’ and ‘class_names’
Return type: Dict[str,Any]

run(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Run transformation

Parameters: model (onnx.ModelProto) – model
Returns: model
Return type: onnx.ModelProto

class vortex.runtime.onnx.graph_ops.embed_class_names_metadata.EmbedClassNamesMetadata(class_names: Union[List[str], Dict])¶

__call__(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Transform model, call derived run function.

Parameters: model (onnx.ModelProto) – model to be transformed
Returns: transformed model
Return type: onnx.ModelProto

classmethod apply(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, class_names: Union[List[str], Dict])¶

Embed class_names information to model

Parameters

model (onnx.ModelProto) – model
class_names (Union[List[str],Dict]) – class_names, either list of string or mapping from string to int

Raises

TypeError – class_names is not dict nor list

Returns

model

Return type

onnx.ModelProto

classmethod parse(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → Dict[int, str]¶

Extract class_names information from model

Parameters: model (onnx.ModelProto) – model
Raises: ValueError – if model doesn’t contains class_labels
Returns: a mapping from class label (int) to class name
Return type: Dict[int,str]

class vortex.runtime.onnx.graph_ops.embed_output_format_metadata.EmbedOutputFormatMetadata(output_format: Dict[str, Union[List[int], int]])¶

__call__(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto¶

Transform model, call derived run function.

Parameters: model (onnx.ModelProto) – model to be transformed
Returns: transformed model
Return type: onnx.ModelProto

classmethod apply(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto, output_format: Dict[str, Union[List[int], int]])¶

Format each ‘output’ in ‘output_format’ and embed it to existing model

Parameters

model (onnx.ModelProto) – model
output_format (Dict[str,Union[List[int],int]]) – output_format, must contains ‘indices’ and ‘axis’ entry

Raises

TypeError – value type of output_format is not dictionary
ValueError – each entry in output_format doesnt contains ‘indices’ and/or ‘axis’

Returns

model with output_format embedded

Return type

onnx.ModelProto

classmethod parse(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto) → Dict[str, Union[List[int], int]]¶

Extract and parse output_format information from model

Parameters: model (onnx.ModelProto) – model
Returns: output_format
Return type: Dict[str,Union[List[int],int]]

run(model: onnx.onnx_ONNX_REL_1_6_ml_pb2.ModelProto)¶

run fn, as required by base class

Parameters: model (onnx.ModelProto) – model to be transformed
Returns: transformed model
Return type: onnx.ModelProto