ConvertArgs

path: Path = Field(default_factory=lambda : ..., exclude=True, description='Path to yaml configuration')

Path to yaml configuration.

model: ModelArgs = Field(default_factory=ModelArgs, description='Model configuration')

Model configuration.

module: ModuleArgs = Field(default_factory=ModuleArgs, description='Module configuration')

Module configuration.

test: TestArgs = Field(default_factory=TestArgs, description='Test configuration')

Test configuration.

transforms: list[TransformSpec] = Field(default_factory=list, description='Transforms configuration')

Transforms configuration.

memory: MemoryArgs = Field(default_factory=MemoryArgs, description='Memory configuration')

Memory configuration.

platform: PlatformArgs = Field(default_factory=PlatformArgs, description='Target platform configuration')

Target platform configuration.

operators: list[OperatorRuleset] = Field(default_factory=list, description='Operator attributes and optimization knob overrides')

Operator attribute rules and optimization knob overrides.

documentation: DocumentationArgs = Field(default_factory=DocumentationArgs, description='Documentation configuration')

Documentation configuration.

optimization: OptimizationArgs = Field(default_factory=OptimizationArgs, description='Optimization goal, approximation gate and knobs')

Optimization goal, approximation gate and knobs.

verbose: int = Field(1, ge=0, le=3, description='Verbosity level')

Verbosity level (0-3).

force: bool = Field(False, description='Force conversion even if output exists')

keep_work_dir: bool = Field(False, description='Keep the temporary work directory and log its path')

Keep the temporary work directory after the conversion instead of removing it, and log its path.

log_file: Path | None = Field(None, description='Optional log file path')

Optional log file path.

operator_rules() -> list[AttributeRuleset]

Return the operator attribute rules, without optimization knobs.

Returns

Returns of operator_rules
Type	Description
`list[AttributeRuleset]`	list[AttributeRuleset]: The rules to resolve operator attributes from.

knob_rules(*, include_model_knobs: bool = True) -> list[AttributeRuleset]

Return the optimization knob rules, resolved like attributes but on their own channel.

The model-wide knobs (optimization.accumulation) become a leading "*" rule, so every operators[].optimization override wins over them with the usual type/id specificity.

Parameters

Parameters of knob_rules
Name	Type	Default	Description
`include_model_knobs`	`bool`	True	Lead with the model-wide knob rule.

Returns

Returns of knob_rules
Type	Description
`list[AttributeRuleset]`	list[AttributeRuleset]: Rules whose attributes are knob values.

model_knob_values() -> dict[str, str]

Return the model-wide value of every knob, auto where unset.

explicit_model_knobs() -> frozenset[str]

Return the knobs set explicitly under optimization, even when set to auto.

operator_knob_rules() -> list[tuple[int, AttributeRuleset]]

Return each operators[] rule that overrides knobs, with its position.

Returns

Returns of operator_knob_rules
Type	Description
`list[tuple[int, AttributeRuleset]]`	list[tuple[int, AttributeRuleset]]: ``(index in operators, rule)``,
`list[tuple[int, AttributeRuleset]]`	where the rule's attributes are its knob overrides.

from_yaml(path: str | Path) -> ConvertArgs

classmethod

Load a conversion config from a YAML file.

The canonical programmatic entry point mirroring the CLI’s --path (without the CLI’s flag-override merge):

config = ConvertArgs.from_yaml(“my-model.yaml”) AotConverter(config=config).convert()

Parameters

Parameters of from_yaml
Name	Type	Default	Description
`path`	`str \| Path`	Required	Path to a YAML mapping of conversion options.

Returns

Returns of from_yaml
Type	Description
`ConvertArgs`	The validated configuration; unknown nested keys warn as usual.

Raises

Errors raised by from_yaml
Type	Description
`FileNotFoundError`	If ``path`` does not exist.
`ValueError`	If the file is not a YAML mapping.