ONNX integration

oneflow.onnx.export(job_func: Callable, model_save_dir: str, onnx_filename: str, continue_on_error: bool = False, opset: Optional[int] = None, extra_opset: Optional[int] = None, shape_override: Optional[Dict[str, List[int]]] = None, external_data: bool = False)

Export a oneflow model into ONNX format.

  • job_func – The job function

  • model_save_dir – The directory containing oneflow model weights. Users are expected to call, wait for the model saving finishing, and pass the argument ‘dir’ as model_save_dir.

  • onnx_filename – a string for the output filename

  • continue_on_error – if an op can’t be processed (aka there is no mapping), continue

  • opset – the opset to be used (int, default is oneflow.python.onnx.constants.PREFERRED_OPSET)

  • extra_opset – list of extra opset’s, for example the opset’s used by custom ops

  • shape_override – dict with inputs that override the shapes given by oneflow

  • external_data – Save weights as ONNX external data, usually to bypass the 2GB file size limit of protobuf.