Repository for RepVGG pruning

This repository is a prunable implementation of RepVGG. We refer to the official RMNet repository.

Requirements

To install requirements:

pip install torch torchvision

Training

To prune the models, run this command:

python train_pruning.py --sr 1e-4 --threshold 5e-4 # sparse training

python train_pruning.py --eval xxx/ckpt.pth # eval

python train_pruning.py --finetune xxx/ckpt.pth # pruned finetuning

TODO: Results

Our model achieves the following performance on :

Contributing

Our code is based on RMNet, RepVGG and nni/amc pruning