Zach

Hi there, I had a look at the config file and I think the most likely culprit is the 'window sampling' parameter. You have chosen `uniform`, which means that the...

Are the images from the same machines? I.e. is the test data acquired in the same way as the training data?

Could you provide some code that will reproduce this bug? What were you running and what was in the `.ini` file you used?

Hi there, I have been looking at this optimizer as well, and thought I'd chime in! Firstly, thanks for the work that you have done. This looks to be a...

1. I would prefer leaving the defaults as they are but also having an `eps_root` argument, defaulting to 0.0. This has the benefit of being a closer match to the...

Hi there, I have posted a new issue in the original repo [here](https://github.com/sail-sg/Adan/issues/12). If they say their experiments were conducted with `no_prox=False`, then I think we can ignore the other...

OK, the authors have responded. Their experiments use `no_prox=False`, the condition that you have not implemented, so I think we do need to implement that one and match their algorithm...

Yes, LGTM. Looking forward to trying it out some more 🙌🏻

Yes, I think that this could be a problem. I agree that it is in a similar boat to the lookahead optimizer.