ml-mobileone
ml-mobileone copied to clipboard
Why latency and model size is smaller (better) with non-reparametrized model in coreml ?
Just found out that, Any intuition behind that?
The coreml models provided are the non-reparametrized right?