v0.6 -- InCoreFalkon, CUDA LAUUM, Bug fixes #10
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- API change to keops options (3 options: "no" keops, "force" keops, "auto" keops) - Center selector should honor device (no more numpy in center selection) - KeOps MMV should be able to run a fully in-core version when presented with CUDA tensors
- Added falkon.to("cuda:0") method
- By converting the model to CUDA, you can call falkon.predict(X) with X a cuda tensor.
- Fixed a number of bugs which were introduced in previous commits
- Add options to manually set the CPU/CUDA cutoff threshold
- Sorted issue #3 for Falkon model (if KeOps is present, data will be coerced to C-contig with a warning) - Converted the preconditioner to work predominantly with tensors. Some operations still need numpy arrays. - Added center selection documentation, and changed naming.
Added CUDA streams to CUDA-TRSM to ensure synchronization
- fix available memory calculation for cuda inputs - fix unexpected ram usage with torch.copy_
- somehow in certain environments importing cublas_gpu succeeds even though there is no CuBLAS available. - the solution is to use torch.cuda.is_available() as a first detection step.
- Install keops - catch modulenotfound error, since this will happen on cpu-only machines - pep8
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Large pull request to incorporate several changes.
The driving change was the implementation of an in-core version of Falkon, suitable for smaller data analyses. Here the data is always kept inside the GPU, thus the model can train much faster. The result is the
InCoreFalkonclass.LAUUM was improved to use a CUDA implementation for the inner-loop function.
Several bug fixes were also introduced, and edge-cases fixed.