Run code without CUDA

First of all, great paper and great that you guys make your code publish available. :-)

For both I need CUDA: train_self_distill.sh and train_mutual_distill.sh. Does the repo/training also works without CUDA available?

Start by reviewing train_self_distill.sh and train_mutual_distill.sh to identify where CUDA is required, then try the repository's training flow in an environment without CUDA. Done means both training scripts can run without CUDA, or the supported limitation is documented clearly.

Written by the indexing model from the issue text.