The notebook contains all the code for creating layer wise causal traces of gpt2 models, as well as some observations from trying out tracing on a variety of prompts.
Outputs have already been produced so there isn't a need to actually run anything if you just want to peruse the traces.
If you do want to run the code, you may need colab pro to load the gpt2-xl model.