IterativeDistillation

Organization :

The main code files are :

• distillation.py which contains the compression algorithms
• models.py for the model architectures
• training.py to train uncomprssed teacher models Other files are mostly test scripts used through the paper.

Usage

To use the code, run the distillation.py file with all needed options (specified in file). Example :

python training.py --dataset cifar10 --datafolder path/to/data --model vgg16 --outfile path/to/model.pt python distillation.py --dataset cifar10 --datafolder path/to/data --teacher_model_file path/to/model.pt

Requirements

• python>=3.7
• pytorch>=1.3.1
• cuda>=9.2
• torchvision>=0.2.2
• tqdm>=4.41.1
• pillow==6.1.0