ONNX

model_navigator.api.onnx

ONNX optimize API.

optimize

optimize(
    model,
    dataloader,
    sample_count=DEFAULT_SAMPLE_COUNT,
    batching=True,
    target_formats=None,
    target_device=DeviceKind.CUDA,
    runners=None,
    optimization_profile=None,
    workspace=None,
    verbose=False,
    debug=False,
    verify_func=None,
    custom_configs=None,
)

Entrypoint for ONNX optimize.

Perform conversion, correctness testing, profiling and model verification.

Parameters:

• model (Union[Path, str]) –

  ONNX model path or string

• dataloader (SizedDataLoader) –

  Sized iterable with data that will be feed to the model

• sample_count (Optional[int], default: DEFAULT_SAMPLE_COUNT ) –

  Limits how many samples will be used from dataloader

• batching (Optional[bool], default: True ) –

  Enable or disable batching on first (index 0) dimension of the model

• target_formats (Optional[Tuple[Union[str, Format], ...]], default: None ) –

  Target model formats for optimize process

• target_device (Optional[DeviceKind], default: CUDA ) –

  Target device for optimize process, default is CUDA

• runners (Optional[Tuple[Union[str, Type[NavigatorRunner]], ...]], default: None ) –

  Use only runners provided as parameter

• optimization_profile (Optional[OptimizationProfile], default: None ) –

  Optimization profile for conversion and profiling

• workspace (Optional[pathlib.Path], default: None ) –

  Workspace where packages will be extracted

• verbose (bool, default: False ) –

  Enable verbose logging

• debug (bool, default: False ) –

  Enable debug logging from commands

• verify_func (Optional[VerifyFunction], default: None ) –

  Function for additional model verification

• custom_configs (Optional[Sequence[CustomConfig]], default: None ) –

  Sequence of CustomConfigs used to control produced artifacts

Returns:

• Package –

  Package descriptor representing created package.

Source code in model_navigator/api/onnx.py

def optimize(
    model: Union[pathlib.Path, str],
    dataloader: SizedDataLoader,
    sample_count: Optional[int] = DEFAULT_SAMPLE_COUNT,
    batching: Optional[bool] = True,
    target_formats: Optional[Tuple[Union[str, Format], ...]] = None,
    target_device: Optional[DeviceKind] = DeviceKind.CUDA,
    runners: Optional[Tuple[Union[str, Type[NavigatorRunner]], ...]] = None,
    optimization_profile: Optional[OptimizationProfile] = None,
    workspace: Optional[pathlib.Path] = None,
    verbose: bool = False,
    debug: bool = False,
    verify_func: Optional[VerifyFunction] = None,
    custom_configs: Optional[Sequence[CustomConfig]] = None,
) -> Package:
    """Entrypoint for ONNX optimize.

    Perform conversion, correctness testing, profiling and model verification.

    Args:
        model: ONNX model path or string
        dataloader: Sized iterable with data that will be feed to the model
        sample_count: Limits how many samples will be used from dataloader
        batching: Enable or disable batching on first (index 0) dimension of the model
        target_formats: Target model formats for optimize process
        target_device: Target device for optimize process, default is CUDA
        runners: Use only runners provided as parameter
        optimization_profile: Optimization profile for conversion and profiling
        workspace: Workspace where packages will be extracted
        verbose: Enable verbose logging
        debug: Enable debug logging from commands
        verify_func: Function for additional model verification
        custom_configs: Sequence of CustomConfigs used to control produced artifacts

    Returns:
        Package descriptor representing created package.
    """
    if isinstance(model, str):
        model = pathlib.Path(model)

    if target_formats is None:
        target_formats = DEFAULT_ONNX_TARGET_FORMATS

    if runners is None:
        runners = default_runners(device_kind=target_device)

    if optimization_profile is None:
        optimization_profile = OptimizationProfile()

    target_formats_enums = enums.parse(target_formats, Format)
    runner_names = enums.parse(runners, lambda runner: runner if isinstance(runner, str) else runner.name())

    if Format.ONNX not in target_formats_enums:
        target_formats_enums = (Format.ONNX,) + target_formats_enums

    config = CommonConfig(
        Framework.ONNX,
        model=model,
        dataloader=dataloader,
        target_formats=target_formats_enums,
        target_device=target_device,
        sample_count=sample_count,
        batch_dim=0 if batching else None,
        runner_names=runner_names,
        optimization_profile=optimization_profile,
        verbose=verbose,
        debug=debug,
        verify_func=verify_func,
        custom_configs=map_custom_configs(custom_configs=custom_configs),
    )

    models_config = ModelConfigBuilder.generate_model_config(
        framework=Framework.ONNX,
        target_formats=target_formats_enums,
        custom_configs=custom_configs,
    )

    builders = [
        preprocessing_builder,
        find_device_max_batch_size_builder,
        tensorrt_conversion_builder,
        correctness_builder,
        performance_builder,
        verify_builder,
    ]

    package = optimize_pipeline(
        model=model,
        workspace=workspace,
        builders=builders,
        config=config,
        models_config=models_config,
    )

    return package