New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

weight量化问题 #101

Open

paranoidai opened this issue Dec 21, 2021 · 0 comments

paranoidai commented Dec 21, 2021

请问在UnsignedQuantizer 里面，对于weights的量化，quant_max_val 为什么会设为(1 << self.bits) - 2呢？对于activation和weights不同处理的目的是？

# for free to join this conversation on GitHub. Already have an account? # to comment