Adopt dynamo cache size to current layer definition #737
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
Adopt setting dynamo cache size to current behavior where number of graphs in compilation in one forward path is equal: number of LlamaDecoderLayer's + 2 (RMSNorm, VocabParallelEmbedding)
anko-intel
requested review from
kzawora-intel,
madamczykhabana,
michalkuligowski,
mgawarkiewicz,
vivekgoe and
afierka-intel
as code owners
anko-intel
changed the title
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Outdated suggestions cannot be applied.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.
Adopt setting dynamo cache size to current behavior where number of graphs in compilation in one forward path is equal:
number of LlamaDecoderLayer's + 2 (RMSNorm, VocabParallelEmbedding)
It is other approach to prepare hot fix after performance regression introduced by vllm-project#11967.
The hot fix #709 restores previous performance results for torch compile mode.
This one partially recover throughput but with big cost of warmup time - as after it much more graphs are compiled during warmup.
Without increasing the cache size torch.compile reach the limits and goes in eager mode which gives low throughput .