https://github.com/enyac-group/single-path-nas

Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours
https://github.com/enyac-group/single-path-nas

Last synced: 25 days ago
JSON representation

Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours

• Host: GitHub
• URL: https://github.com/enyac-group/single-path-nas
• Owner: enyac-group
• License: apache-2.0
• Created: 2019-03-29T17:56:32.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-12-14T08:08:59.000Z (almost 4 years ago)
• Last Synced: 2024-05-19T00:27:11.303Z (7 months ago)
• Language: Python
• Homepage: https://arxiv.org/abs/1904.02877
• Size: 175 KB
• Stars: 397
• Watchers: 15
• Forks: 61
• Open Issues: 16
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-AutoML-and-Lightweight-Models - dstamoulis/single-path-nas

README

        
# Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours

## Requirements

* Access to Cloud TPUs ([Official Cloud TPU Tutorial](https://cloud.google.com/tpu/docs/tutorials/mnasnet))

* Tensorflow 1.12

* Python 3.5+

## Updates

* 04/05/19: Single-Path NAS search code released.

## Contents

* NAS Search [nas-search](/nas-search/): Employ NAS search 

* Runtime Modeling [runtime-modeling](/runtime-modeling/): Generate ConvNets to profile

* Train ConvNet [train-final](/train-final/): Fully train found ConvNet on ImageNet

## Citation

Please cite the Single-Path paper ([link](https://arxiv.org/abs/1904.02877)) 

in your publications if this repo helps your research:

    @inproceedings{stamoulis2019singlepath,

      author = {Stamoulis, Dimitrios and Ding, Ruizhou and Wang, Di and Lymberopoulos, Dimitrios and Priyantha, Bodhi and Liu, Jie and Marculescu, Diana}

      booktitle = {arXiv preprint arXiv:1904.02877},

      title = {Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours},

      year = {2019}

    }