Skip to content
forked from ricky40403/DSQ

pytorch implementation of "Differentiable Soft Quantization: Bridging Full-Precision and Low-Bit Neural Networks"

Notifications You must be signed in to change notification settings

YumingChang02/DSQ

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 

Repository files navigation

DSQ

pytorch implementation of "Differentiable Soft Quantization: Bridging Full-Precision and Low-Bit Neural Networks"


The Origin Paper : https://arxiv.org/abs/1908.05033


This repository follow the Algorithm 1 in the paper.
As mention in the paper:

For clipping value l and u, we try the following two strategies:  
moving average statistics and optimization by backward propagation.

Because the paper fine-tunes from the pre-trained model, so it can find the clipping value.
Instead of using the strategies, this repository uses the max value of int32 as the initial value.
It should not affect the value range (because the parameter of the deep model should not too large), and most of the edge device range is up to int32.

Note

note that this has not been test yet

About

pytorch implementation of "Differentiable Soft Quantization: Bridging Full-Precision and Low-Bit Neural Networks"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%