Adding dynamic swap number and defragmenter warmup #183
Conversation
This file contains hidden or 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
Signed-off-by: Krzysztof Smusz <ksmusz@habana.ai>
Signed-off-by: Krzysztof Smusz <ksmusz@habana.ai>
Signed-off-by: Krzysztof Smusz <ksmusz@habana.ai>
ksmusz
requested review from
adobrzyn,
kzawora-intel,
mswiniarsk and
xuechendi
as code owners
Contributor
Author
|
/run-gaudi-tests |
Contributor
Author
|
/run-gaudi-tests |
madamczyk-intel
approved these changes
Sep 19, 2025
mswiniarsk
approved these changes
Sep 19, 2025
Contributor
Author
|
/run-gaudi-tests |
mswiniarsk
added a commit
that referenced
this pull request
Sep 19, 2025
Introducing dynamic swap buckets to defragmenter, together with defragmenter warmup. Currently only a maximum of 32 blocks can be swapped of one iteration of a defragmenter. This change introduces a bucketing system, which asserts the minimal size bucket of swaps to be done in current defragmenter iteration based on actual number of blocks, that need to be swapped. Size of the buckets range from 8 swaps up to 512 swaps in a single defragmenter run. As the number of possible swap buckets grew from a single size bucket, a warmup of defragmenter has been added. Thanks to the warmup, no additional graph compilations connected to the defragmenter were visible during the inference. cherry-pick #183 Signed-off-by: Krzysztof Smusz <ksmusz@habana.ai> Co-authored-by: Marcin Swiniarski <marcin.swiniarski@intel.com>
Contributor
Author
|
/run-gaudi-tests |
2 similar comments
Contributor
Author
|
/run-gaudi-tests |
Contributor
Author
|
/run-gaudi-tests |
mswiniarsk
requested review from
afierka-intel and
mgawarkiewicz-intel
as code owners
Contributor
Author
|
/run-gaudi-tests |
Contributor
Author
|
/run-gaudi-tests |
1 similar comment
Contributor
Author
|
/run-gaudi-tests |
iboiko-habana
pushed a commit
to iboiko-habana/vllm-gaudi
that referenced
this pull request
Oct 2, 2025
Introducing dynamic swap buckets to defragmenter, together with defragmenter warmup. Currently only a maximum of 32 blocks can be swapped of one iteration of a defragmenter. This change introduces a bucketing system, which asserts the minimal size bucket of swaps to be done in current defragmenter iteration based on actual number of blocks, that need to be swapped. Size of the buckets range from 8 swaps up to 512 swaps in a single defragmenter run. As the number of possible swap buckets grew from a single size bucket, a warmup of defragmenter has been added. Thanks to the warmup, no additional graph compilations connected to the defragmenter were visible during the inference. --------- Signed-off-by: Krzysztof Smusz <ksmusz@habana.ai> Co-authored-by: Marcin Swiniarski <marcin.swiniarski@intel.com> Signed-off-by: Iryna Boiko <iboiko@habana.ai>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Introducing dynamic swap buckets to defragmenter, together with defragmenter warmup.
Currently only a maximum of 32 blocks can be swapped of one iteration of a defragmenter. This change introduces a bucketing system, which asserts the minimal size bucket of swaps to be done in current defragmenter iteration based on actual number of blocks, that need to be swapped. Size of the buckets range from 8 swaps up to 512 swaps in a single defragmenter run.
As the number of possible swap buckets grew from a single size bucket, a warmup of defragmenter has been added. Thanks to the warmup, no additional graph compilations connected to the defragmenter were visible during the inference.