Max Balandat
Max Balandat
> I did initially try and come up with something to contribute to GPyTorch and/or linear_operator, but it was harder than anticipated to make it compatible and not introduce slowdowns...
> should I submit a PR for this, or is there an obvious reason that it wouldn't work in other use cases? A PR for this would be great -...
cc @sdaulton, @jandylin, @SebastianAment re excessive memory usage in Kronecker MTGPs
cc @JonathanWenger
Thanks for raising this. Yes this is the same reason as for #2550 - the `.to()` doesn't move the attributes over the GPU. @hvarfner in case you have any immediate...
Yeah, I understand this isn't great behavior. > Maybe something like this could be implemented inside BoTorch (when fitting a model or optimizing an acquisition function). This would be an...