This repository has been archived by the owner on Dec 18, 2023. It is now read-only.
Replace numpy cholesky with torch.linalg.cholesky_ex #996
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
In the past we used
numpy.cholesky
in NMC implementation becausetorch.cholesky
fails very slowly in comparison. Since a workaround has been proposed in pytorch/pytorch#34272, we can now switch back to using torch and avoid the numpy conversion.A small benchmark to show the run time difference with the workaround: N1136967
(I just happen to spot this when I was looking at NMC to see how to merge the infra together. There will probably be quite a few changes coming on single site infra -- I will try to keep each of them small to make reviewing easier :).)
Differential Revision: D30895231