Intro
ML.NET supports exporting trained models to the ONNX file format, but only allows export to the latest version/opset. The version of the exported model can also change as ML.NET is updated, which can break compatibility with other toolsets that only support older ONNX versions.
Feature Request
- Support exporting models to older ONNX versions and opsets (ONNX version 1.4, opset 9 or older)
Use case
We would like to run a model trained with ML.NET on Windows devices using the inbox Windows ML runtime, but it does not support the latest ONNX versions which ML.NET exports to.
Workarounds
- Exported models can be converted using the ONNX Version Converter, but it does not work on all models.
- Models can also be hand-edited.
Intro
ML.NET supports exporting trained models to the ONNX file format, but only allows export to the latest version/opset. The version of the exported model can also change as ML.NET is updated, which can break compatibility with other toolsets that only support older ONNX versions.
Feature Request
Use case
We would like to run a model trained with ML.NET on Windows devices using the inbox Windows ML runtime, but it does not support the latest ONNX versions which ML.NET exports to.
Workarounds