thanks for linking.

Yes the paper compares the new architecture (that is also a fork of my implementation of nanoGPT) with Karpathy's nanoGPT. There are also links to the code and bench used.