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

Requirements

• Python 3.5+
• CUDA 9.0
• NCCL 2.3.7
• Conda 4.6.12

Quick Start

• Create and activate the conda environment:

  conda env create -f environment.yml
  source activate sp-nas

• Add environment variables:

  export LD_LIBRARY_PATH=/usr/local/nccl-2.3.7/lib:/usr/local/cuda-9.0/lib64
  export PATH=/usr/local/cuda-9.0/bin:$PATH

• NAS Search nas-search: Employ NAS search
• Train ConvNet train-final: Fully train found ConvNet on ImageNet

Citation

Please cite the Single-Path paper (link) 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}
}