Quantization in Core ML Tools

Hi,


In the WWDC 18 Talk about "Whats new in Core ML part 2" it is mentioned that it is possible to lower your Core ML model size by using quantization to reduce the precision of the network weights. I can't find any meniton of how to do this on the Core ML Tools website. I can only find https://developer.apple.com/documentation/coreml/reducing_the_size_of_your_core_ml_app, but this is not what was show in the talk. Does anyone have an example of how to do this?



Kind regards,


Anders Sejersbøl.

Replies

It appears that the coremltools repo on GitHub has not been updated yet with the new features.

OK, thanks for the clarification.