transformers.py
contains the code to train the model. Grokking_Analysis.ipynb
contains the code to load the saved checkpoints for the mainline run, calculate the progress metrics on it, and plots the figures. Non_Modular_Addition_Grokking_Tasks.ipynb
contains training code for the non-modular addition experiments.
michahu / progress-measures-paper Goto Github PK
View Code? Open in Web Editor NEWThis project forked from mechanistic-interpretability-grokking/progress-measures-paper