Open notebook_examples/memorypro_ntk_quickstart.ipynb and run all cells.
The notebook requires a CUDA-enabled PyTorch installation; its saved outputs are
also embedded for immediate viewing.
The bundle contains the matrix-free kpflow source and the minimum saved data
needed for the default-initialized and five-fixed-point vanilla-RNN comparison.