$ conda create -n [envname] python==3.10.14
$ conda activate [envname]
$ conda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=12.1 -c pytorch -c nvidia
$ pip install -r requirements.txt

Here i use cifar-10, and use fp16 to accelerate

$ python run.py -data ./datasets -dataset-name cifar10 -j 4 --log-every-n-steps 100 --epochs 100 --batch-size 256 --fp16-precision

If you want to run it on CPU (for debugging purposes) use the --disable-cuda option.

change the checkpoint_path, then run the feature_test.py

Some training and testing records can be found in running folder