onnx_diagnostic.torch_onnx.runtime_info

class onnx_diagnostic.torch_onnx.runtime_info.RuntimeDevice(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source][source]

Device definition

class onnx_diagnostic.torch_onnx.runtime_info.RuntimeValue(name: str, dtype: Any | None = None, shape: Tuple[str | int, ...] | None = None, value: Any | None = None, first_used: int | None = None, last_used: int | None = None, created: int | None = None, is_shape: bool | None = None, kind: RuntimeValueKind | None = None, device: RuntimeDevice | None = None)[source][source]

Describes a value used during the execution of a model.

clean_value()[source][source]

Sets value to None.

property has_value: bool

Tells if value is specified.

property is_initializer: bool

Tells if it is an initializer.

property is_input: bool

Tells if it is an input.

property is_output: bool

Tells if it is an output.

set_value(value: Tensor | TensorLike)[source][source]

Sets the value.

string_type() → str[source][source]

Returns a string describing the value.

class onnx_diagnostic.torch_onnx.runtime_info.RuntimeValueKind(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source][source]

Kind of result.

onnx_diagnostic.torch_onnx.runtime_info.first_used_last_used(proto: FunctionProto | GraphProto | ModelProto, constant_as_initializer: bool = False) → Dict[str, RuntimeValue][source][source]

Builds first used, last used information for every result in the model.

Parameters:

  • proto – model, graph or function

  • constant_as_initializer – outputs of node Constant is tagged as INITIALIZER

Returns:

dictionary of RuntimeValue

onnx_diagnostic.torch_onnx.runtime_info.get_hidden_inputs(graph: GraphProto) → Set[str][source][source]

Returns the hidden inputs (inputs coming from an upper context) used by a subgraph.

onnx_diagnostic.torch_onnx.runtime_info.set_is_shape(node: NodeProto, values: Dict[str, RuntimeValue], drop: Set[str] | None = None) → List[str][source][source]

Sets attribute is_shape for outputs of a node.

Parameters:

  • node – node to process

  • values – stored results, values in this dictionary are updated

  • drop – variables not to consider because the come from the graph holding this subgraph

Returns:

list of modified results