Thank you for such a cool project.

If I want to migrate from AdamW (or other optimizers), what is the rule of thumb of adjusting the hyperparameters (e.g, learning rate, weight decay, etc.)?

This information would be useful to include in the main README as well!