Niccolò Ajroldi

Is there any update regarding this issue? Avoid calibration doesn't seem to be a solution, since estimated variance can be negative. Thank you in advance!

Sure! Happy to help! On Wed, Mar 1, 2023, 12:16 ArtsiomWB ***@***.***> wrote: > Hi @Niccolo-Ajroldi ! thank you for > reporting this. I have submitted this to our eng...

@LeoXinhaoLee This might help: https://github.com/state-spaces/mamba/issues/184

Hi @priyakasimbeg, thanks for looking into this! I'll explain better. In classical weight averaging (Polyak Averaging, SWA, LAWA...), we collect a copy of model_parameters once in a while, aggregate those...

Notice that the pull request modifies current submissions adding the extra argument `is_eval_step`, but this is not really necessary. We can make the code backward compatible by putting the call...

Sure, happy to join the WG meeting on Thursday. Making a step longer is indeed a good solution, thanks! Regarding the problem of a too long final step, we could...