mrlj-hash
Results
2
comments of
mrlj-hash
I don't have the hardware to test this myself, but does the same issue happen for _single-output_ non-exact GP models when using multiple GPUs?
Has there been more thought on this @gpleiss since the issue was opened? There are recent developments (https://arxiv.org/abs/1910.10596) on this orthogonal decoupling approach which allow greater flexibility that would be...