Python API¶

This document lists the application programming interface (API) that the generated to Python. With these APIs, applications can:

upload a model to the ONNC bench SaaS,

compile the neural network, and

download the compiled model from ONNC bench SaaS

Workspace¶

Workspace class represents a workspace in the remote service.

from onnc.bench import Workspace;

Upload A Model¶

upload_model creates a remote workspace in ONNC bench SaaS and upload the model into the workspace.

onnc.bench.Workspace.upload_model(model,
                                  samples,
                                  input_name = None,
                                  output_name = None,
                                  input_as_nchw: str = "auto")

model: Union[str, object]

A file or a directory path to a serialized model file, or a neurl network model object.

samples: Union[str, object]

A file or a directory path to a serialized Numpy dataset, or a Numpy dataset object.

input_name: str

The input tensor name of the given model.

output_name: str

The output tensor name of the given model.

input_as_nchw: str

This parameter is used to define the input format. It should be one of the noset, auto, or as_input_name.

return

A python dict object contains success, model_id, and sample_id.

example

{
  'success': True,
  'model_id': '',
  'sample_id': ''
}

Compile The Neural Network¶

compile asks ONNC bench SaaS to transform the given model into C function calls.

onnc.bench.Workspace.compile(board, ram_size=0)

board: str

The name of the supported SoC board, for example, NUMAKER_IOT_M487.

ram_size: int

Limit of free RAM size in the board. The value depends on the environmental conditions, such as OS, application.

return

A python dict object contains success, loadable_id, board, ram_size, and report.

example

{
  'success': success,
  'loadable_id': loadable_id,
  'board': board,
  'ram_size': ram_size,
  'report': {'ram': ram, 'rom': rom}
}

Download The Compiled Model¶

download downloads the produced C function calls from ONNC bench SaaS.

onnc.bench.Workspace.download(download_path, unzip=True)

download_path: str: The path to save the file.
unzip: bool: Unzip the results when true.