Parallel sampling with threadpool #1252
Open
+271
−148
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
post rebase adjustments fix finish iteration move currently_processed_tokens update switch to async experimental threadpool remove access to shared struct in parallelized code synchronize beam search part
github-actions
bot
added
category: continuous batching
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Outdated suggestions cannot be applied.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 PR implements the same functionality as: #1233, but in a different manner. Only one of them should be merged.
Since pipeline logic is executed on a single thread, there are periods of low CPU usage while pipeline is not executing inference, but some other logic like sampling which can take quite large fraction of time. Currently after inference is done we sample from each sequence group in a loop on a single thread which becomes an issue with sampling parameters that significantly extend sampling time for a single sequence group.
This PR extracts sampling logic for single sequence group into a separate method that can be executed independently from any other sequence group. In includes generic thread pool implementation that spawns certain amount of threads that are used to run sampling logic for different sequence groups in parallel.
Performance measurements confirm improvement especially for non greedy sampling and with high concurrency (the more sequence groups are scheduled for inference the more benefit from parallel sampling).