enable feature score data collection in torchrec #3285
+76
−2
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
meta-cla
bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D79864431 |
|
This pull request was exported from Phabricator. Differential Revision: D79864431 |
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Aug 15, 2025
Summary: Pull Request resolved: pytorch#3285 Add enable_feature_score_weight_accumulation flag to ShardedEmbeddingCollection. When this flag is true, and dedup ec index is true, we'll accumulate kjt weight and count and reset back to kjt weight, to allow input dist to distribute feature score. this change is part of ZCH v.Next feature score eviction story: - collect score for every feature id in model, e.g. for positive id set to 0.5, and negative id set to 0.2. - set score as the input id list feature kjt's weight value - in EC forward, if there is ID dedup, aggregate the id score and occurrence of each id. - distribute the id score in kjt weight - in KVZCH embedding kernel, call forward with weight as an optional parameter in ZCH TBE backend (separate diffs): - set the feature score to ZCH TBE backend - run eviction based on the id score value for the whole story, please reference here: https://docs.google.com/document/d/1TJHKvO1m3-5tYAKZGhacXnGk7iCNAzz7wQlrFbX_LDI/edit?tab=t.0 Differential Revision: D79864431
emlin
force-pushed
the
export-D79864431
branch
from
999a168 to
7cfb966
Compare
Differential Revision: D79591336
|
This pull request was exported from Phabricator. Differential Revision: D79864431 |
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Aug 15, 2025
Summary: Pull Request resolved: pytorch#3285 Add enable_feature_score_weight_accumulation flag to ShardedEmbeddingCollection. When this flag is true, and dedup ec index is true, we'll accumulate kjt weight and count and reset back to kjt weight, to allow input dist to distribute feature score. this change is part of ZCH v.Next feature score eviction story: - collect score for every feature id in model, e.g. for positive id set to 0.5, and negative id set to 0.2. - set score as the input id list feature kjt's weight value - in EC forward, if there is ID dedup, aggregate the id score and occurrence of each id. - distribute the id score in kjt weight - in KVZCH embedding kernel, call forward with weight as an optional parameter in ZCH TBE backend (separate diffs): - set the feature score to ZCH TBE backend - run eviction based on the id score value for the whole story, please reference here: https://docs.google.com/document/d/1TJHKvO1m3-5tYAKZGhacXnGk7iCNAzz7wQlrFbX_LDI/edit?tab=t.0 Reviewed By: duduyi2013 Differential Revision: D79864431
emlin
force-pushed
the
export-D79864431
branch
from
7cfb966 to
599350c
Compare
Summary: Pull Request resolved: pytorch#3285 Add enable_feature_score_weight_accumulation flag to ShardedEmbeddingCollection. When this flag is true, and dedup ec index is true, we'll accumulate kjt weight and count and reset back to kjt weight, to allow input dist to distribute feature score. this change is part of ZCH v.Next feature score eviction story: - collect score for every feature id in model, e.g. for positive id set to 0.5, and negative id set to 0.2. - set score as the input id list feature kjt's weight value - in EC forward, if there is ID dedup, aggregate the id score and occurrence of each id. - distribute the id score in kjt weight - in KVZCH embedding kernel, call forward with weight as an optional parameter in ZCH TBE backend (separate diffs): - set the feature score to ZCH TBE backend - run eviction based on the id score value for the whole story, please reference here: https://docs.google.com/document/d/1TJHKvO1m3-5tYAKZGhacXnGk7iCNAzz7wQlrFbX_LDI/edit?tab=t.0 Reviewed By: duduyi2013 Differential Revision: D79864431
|
This pull request was exported from Phabricator. Differential Revision: D79864431 |
emlin
force-pushed
the
export-D79864431
branch
from
599350c to
18d7617
Compare
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.
Summary:
Add enable_feature_score_weight_accumulation flag to ShardedEmbeddingCollection. When this flag is true, and dedup ec index is true, we'll accumulate kjt weight and count and reset back to kjt weight, to allow input dist to distribute feature score.
this change is part of ZCH v.Next feature score eviction story:
in ZCH TBE backend (separate diffs):
for the whole story, please reference here:
https://docs.google.com/document/d/1TJHKvO1m3-5tYAKZGhacXnGk7iCNAzz7wQlrFbX_LDI/edit?tab=t.0
Differential Revision: D79864431