Giters

huawei-noah / HEBO

Bayesian optimisation & Reinforcement Learning library developped by Huawei Noah's Ark Lab

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CUDA out of memory for BOiLS

Flians opened this issue · comments

commented

X1_full = torch.repeat_interleave(X1.unsqueeze(0), len(indicies), dim=0)[ RuntimeError: CUDA out of memory. Tried to allocate 3.89 GiB (GPU 0; 11.93 GiB total capacity; 3.49 GiB already allocated; 3.65 GiB free; 7.67 GiB reserved in total by PyTorch)

BOiLS takes up too many GPU memories. Do you have any suggestions about it?

commented

Hi! This is indeed a drawback of the current SSK implementation. We are actually working on this aspect to make SSK computation less memory intensive. For now the most direct fix would be to use GP with inducing points when the number of training points is too large for the exact GP to be fit on a GPU.