Samuel Müller
Samuel Müller
Hi @rwightman, Thanks for the long answer. I saw your fixes in the pipeline when I used it to run EffNet trainings and found it interesting you came up with...
Hi @rwightman, Thanks for the second reply as well. A quick update on the requested experiments: I am right now training a big model: EfficientNet-V2-M with hyper-parameters inspired by your...
@rwightman Ok, good! Oh yes, of course! I should know that :D The first runs of V2-M are done. My config diff for the RA version on 32 workers is...
Wow this is open from almost a year ago... I think someone could get a lot citations / clicks if they did a proper benchmark of transformer train/inference across platforms...
That can very well be. The GP prior is not very helpful for most datasets, and the MLP prior does include the SCM setting if I remember correctly, so that...
does this issue persist?
I know this is all very sub-optimally written for people working on the code now, as this is very much still the code in which we tried out a lot...
Ok, I updated our README to reflect that it does not work for py>=3.12, did you work on some code to fix the dependency issue? Otherwise, I think I can...
Oh, very old issue. I have fixed this now after moving the new model into this repo.
The weird usages are usages with more than three dimension in the initial conditions. They are inside the functions listed above. You can quickly try it out by printing the...