operator

OpIdattribute
AirOperatorSourceclassThe model-file operator an AIR operator was parsed from.
AirOperatorclassIntermediate Representation of an operator (node) in the graph.

OpId = str

AirOperatorSource()

The model-file operator an AIR operator was parsed from.

Provisional: importable, but outside the compatibility promise; a minor release can change or remove it.

name: str = Field(..., description="The operator's name in the model file")

The operator’s name in the model file: a LiteRT builtin operator name such as LESS, or a custom operator’s canonical key.

op_type: str = Field(..., description='Canonical key of the AIR type the parser gave it')

The canonical key of the AIR type the parser gave it, which a transform that rewrites the operator leaves behind.

indices: tuple[int, ...] = Field(..., description="Operator indices in the model file's subgraph")

The operator’s index, or the indices a LiteRT model hook states for an operator it created in place of others, in the converted subgraph of the model file; empty for an operator a hook created without stating any.

AirOperator()

Intermediate Representation of an operator (node) in the graph.

id: OpId = Field(..., description="Unique node identifier: a decimal integer string from 0 to 2147483647 (e.g. '3')")

Unique node identifier: a decimal integer string from 0 to 2147483647 that no other operator in the model has (e.g. ‘3’).

op_type: AirOpType = Field(..., description='The operator unique type.')

The operator unique type.

input_ids: list[TensorId] = Field(default_factory=list, description='Input tensor IDs.')

Input tensor IDs.

output_ids: list[TensorId] = Field(default_factory=list, description='Output tensor IDs.')

Output tensor IDs.

named_tensors: dict[str, TensorId] = Field(default_factory=dict, description='Local tensor names to IDs mapping.')

Local tensor names to IDs mapping.

options: AirOperatorOptions = Field(default_factory=AirOperatorOptions, description='Typed options/config for this op.')

Typed options/config for this op.

source: AirOperatorSource | None = Field(None, description='The model-file operator it was parsed from; None for one a transform created.')

The model-file operator it was parsed from, which the model loader records; None for an operator a transform created.

tensor_ids: list[TensorId]

All tensor IDs (inputs + outputs + named tensors) associated with this operator.

fan_in: int

How many input tensors this operator consumes.

fan_out: int

How many output tensors this operator produces.

print_info(verbose: int = 0)

Debug print the template values for the operator.