I’ve gone into pretty great detail on the visualization in the README of my repo. The main utility is detecting individual layers being overfit.

There are some specifics about OpenPirate that I’m not at liberty to share at the moment, but those are unrelated to this visualization. I’ve published the model weights under a permissive license, and I hope to publish more of the training code in the future.

If you have any questions about how to use the code in my neural flow repo just ask.