Skip to content

Comment on PyTorch Mobile

Comments

Yes yes yes! I've been anticipating this for so long.

However I'm a bit skeptical about doing quantization after training, in my experience you have to do quantization-aware training for there not be a large performance decrease. I guess it works though otherwise they wouldn't have released it?

According to their docs[1], three different ways of quantizing your model are supported - one of which is quantization-aware training (3).

[1] - https://pytorch.org/docs/master/quantization.html#quantizati...

https://arxiv.org/abs/1906.04721 is an example paper where they perform data-free quantization after training (using batchnorm parameters to get information about the data distribution) without large performance drops.

i hate microsoft's way of getting things done! stop reinventing the wheel !

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.