Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/bzantium/pytorch-admm-pruning
Prune DNN using Alternating Direction Method of Multipliers (ADMM)
https://github.com/bzantium/pytorch-admm-pruning
admm deep-neural-networks pruning
Last synced: about 7 hours ago
JSON representation
Prune DNN using Alternating Direction Method of Multipliers (ADMM)
- Host: GitHub
- URL: https://github.com/bzantium/pytorch-admm-pruning
- Owner: bzantium
- License: bsd-3-clause
- Created: 2019-08-21T08:29:59.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2019-10-15T03:35:26.000Z (almost 5 years ago)
- Last Synced: 2023-03-11T04:36:43.609Z (over 1 year ago)
- Topics: admm, deep-neural-networks, pruning
- Language: Python
- Homepage:
- Size: 15.6 KB
- Stars: 80
- Watchers: 4
- Forks: 17
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# pytorch-admm-prunning
It is a pytorch implementation of DNN weight prunning with ADMM described in [**A Systematic DNN Weight Pruning Framework using Alternating Direction Method of Multipliers**](https://arxiv.org/abs/1804.03294).
## _Train and test_
- You can simply run code by
```
$ python main.py
```
- In the paper, authors use **l2-norm regularization** so you can easily add by
```
$ python main.py --l2
```
- Beyond this paper, if you don't want to use _predefined prunning ratio_, admm with **l1 norm regularization** can give a great solution and can be simply tested by
```
$ python main.py --l1
```
- There are two dataset you can test in this code: **[mnist, cifar10]**. Default setting is mnist, you can change dataset by
```
$ python main.py --dataset cifar10
```
## _Models_
- In this code, there are two models: **[LeNet, AlexNet]**. I use LeNet for mnist, AlexNet for cifar10 by default.
## _Optimizer_
- To prevent prunned weights from updated by optimizer, I modified Adam (named PruneAdam).
## _References_
For this repository, I refer to _[KaiqiZhang's tensorflow implementation](https://github.com/KaiqiZhang/admm-pruning)_.