flytortoise / ccep

This package includes the Python code of the paper 'Neural Network Pruning by Cooperative Coevolution'.

ATTN: This package is free for academic usage. You can run it at your own risk. For other purposes, please contact Dr. Chao Qian ([email protected]).

ATTN2: ATTN2: This package was developed by Mr. Haopu Shang ([email protected]) and Mr. Jia-Liang Wu ([email protected]). For any problem concerning the code, please feel free to contact Mr. Shang or Mr. Wu.

Requirements

• python==3.8.3
• torch==1.8.2
• torchvision==0.9.2
• thop==0.0.31
• prefetch_generator==1.0.1

Usage

Pruning on CIFAR-10

# ResNet56
python pruning.py --dataset cifar10 --arch resnet56  \
                  --ft_epoch 100 --lr_milestone 50 \
                  --dict_path ./models/resnet56.th  --pop_init_rate 0.95  \
                  --prune_limitation 0.90 --batch-size 128 --valid_ratio 0.8  \
                  --run_epoch 20
# VGG16
python pruning.py --dataset cifar10 --arch vgg  \
                  --ft_epoch 100 --lr_milestone 50 \
                  --dict_path ./models/vgg16.th  --pop_init_rate 0.95  \
                  --prune_limitation 0.90 --batch-size 128 --valid_ratio 0.8  \
                  --run_epoch 20

Pruning On ImageNet

# ResNet-50
python pruning.py --dataset ImageNet --arch resnet50  \
                  --ft_epoch 60 --lr_milestone 20 40 50 \
                  --data ./data/ImageNet  --pop_init_rate 0.9  \
                  --prune_limitation 0.85 --batch-size 256 --valid_ratio 0.99  \
                  --run_epoch 20
# ResNet-34
python pruning.py --dataset ImageNet --arch resnet34  \
                  --ft_epoch 60 --lr_milestone 20 40 50 \
                  --data ./data/ImageNet  --pop_init_rate 0.9  \
                  --prune_limitation 0.85 --batch-size 256 --valid_ratio 0.99  \
                  --run_epoch 20

You can find all the commands in the file 'run.sh' .