Need to convert to .ort format about scikit_learn_android_demo HOT 3 CLOSED

@skottmckay The blog and Colab notebook have been updated with the information you've provided on ONNX Runtime. Thanks for improving the project!

from scikit_learn_android_demo.

Thanks for the information @skottmckay! Does ONNX Runtime model provide any additional optimization over the vanilla ONNX model?

from scikit_learn_android_demo.

Its main benefit is allowing usage of the smaller build (onnxruntime-mobile android package) if binary size is a big concern. Same graph optimizations are done by ONNX Runtime for either model format - they're just done ahead of time when creating the .ort format model so we don't have to include optimizer code in the smaller build.

However, there's a chance your model might not be supported by the onnxruntime-mobile package due to the operators/types it uses, so using a .onnx model with onnxruntime-android is the simplest and safest option.

If binary size is an issue and your model is not supported by onnxruntime-mobile, you can do a custom build to create a package that only includes the operators and types your model requires. This provides the smallest possible binary size.

from scikit_learn_android_demo.