use RAFT for select_k on the gpu #656
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
This changes to use RAFT https://github.com/rapidsai/raft for GPU top-k code instead of faiss. The RAFT version is quite a bit faster, and also doesn't have the same performance issues with small batch sizes that faiss has (meaning we can delete a bunch of code that was trying to work around that).
Use the raft cublas handle and thrust policy so that we use the stream associated with the raft handle (rather than just the default stream)
raft::select_k doesn't always return sorted outputs - handle this in unittest
|
Benchmarking this change on a dataset of github stars - containing 9M items with 96 dimensional embeddings, shows a good improvement in queries per second:
|
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Outdated suggestions cannot be applied.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.
This changes to use RAFT https://github.com/rapidsai/raft for GPU top-k code instead of faiss. The RAFT version is quite a bit faster, and also doesn't have the same performance issues with small batch sizes that faiss has (meaning we can delete a bunch of code that was trying to work around that). RAFT also doesn't have limitations on the size of K, where faiss is limited to k less than 2048.