How to quantize weights only on Imagenet?

Open sachitkuhar opened this issue 5 years ago • 2 comments

I wanted to quantize weights only. Should 'qa' be set equal to 0/32? Thanks

Aug 11 '20 11:08 sachitkuhar

You could set 'qa' to 0 to train a cifar10 model with weight quantization only. But we didn't provide the interface to quantize weight only for ImageNet. If you want, you should change related codes in learned_quantization.py

Aug 22 '20 02:08 EowinYe

Thanks for your reply!

In cifar10-vgg-small.py- if 'qa'=0 then you do not use 'QuantizedActiv' function. Does this imply that if 'QuantizedActiv' function is an identity function; Imagenet.py will perform weight quantization only?

Thanks!

Aug 22 '20 07:08 sachitkuhar