add auto feature score collection to EC #5030
Open
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
Summary: X-link: meta-pytorch/torchrec#3474 X-link: facebookresearch/FBGEMM#2043 Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping. If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id. Here is the sample eviction policy config in embedding_table config to enable auto score collection: virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy( training_id_eviction_trigger_count=260_000_000, # 260M training_id_keep_count=160_000_000, # 160M enable_auto_feature_score_collection=True, feature_score_mapping={ "sparse_public_original_content_creator": 1.0, }, feature_score_default_value=0.5, ), Differential Revision: D83945722
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Oct 20, 2025
Summary: X-link: pytorch/FBGEMM#5030 X-link: facebookresearch/FBGEMM#2043 Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping. If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id. Here is the sample eviction policy config in embedding_table config to enable auto score collection: virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy( training_id_eviction_trigger_count=260_000_000, # 260M training_id_keep_count=160_000_000, # 160M enable_auto_feature_score_collection=True, feature_score_mapping={ "sparse_public_original_content_creator": 1.0, }, feature_score_default_value=0.5, ), Additionally the counter collected previously during EC dedup is not used by kvzch backend, so this diff removed that counter and allow KJT to transfer a single float32 weight tensor to backend. This allows feature score collection for EBC since there could have another float weight for EBC pooling already. Differential Revision: D83945722
emlin
added a commit
to emlin/FBGEMM
that referenced
this pull request
Oct 20, 2025
Summary: X-link: meta-pytorch/torchrec#3474 X-link: facebookresearch/FBGEMM#2043 Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping. If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id. Here is the sample eviction policy config in embedding_table config to enable auto score collection: virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy( training_id_eviction_trigger_count=260_000_000, # 260M training_id_keep_count=160_000_000, # 160M enable_auto_feature_score_collection=True, feature_score_mapping={ "sparse_public_original_content_creator": 1.0, }, feature_score_default_value=0.5, ), Additionally the counter collected previously during EC dedup is not used by kvzch backend, so this diff removed that counter and allow KJT to transfer a single float32 weight tensor to backend. This allows feature score collection for EBC since there could have another float weight for EBC pooling already. Differential Revision: D83945722
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Oct 21, 2025
Summary: X-link: pytorch/FBGEMM#5030 X-link: facebookresearch/FBGEMM#2043 Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping. If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id. Here is the sample eviction policy config in embedding_table config to enable auto score collection: virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy( training_id_eviction_trigger_count=260_000_000, # 260M training_id_keep_count=160_000_000, # 160M enable_auto_feature_score_collection=True, feature_score_mapping={ "sparse_public_original_content_creator": 1.0, }, feature_score_default_value=0.5, ), Additionally the counter collected previously during EC dedup is not used by kvzch backend, so this diff removed that counter and allow KJT to transfer a single float32 weight tensor to backend. This allows feature score collection for EBC since there could have another float weight for EBC pooling already. Reviewed By: EddyLXJ Differential Revision: D83945722
emlin
added a commit
to emlin/FBGEMM
that referenced
this pull request
Oct 21, 2025
Summary: X-link: meta-pytorch/torchrec#3474 X-link: facebookresearch/FBGEMM#2043 Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping. If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id. Here is the sample eviction policy config in embedding_table config to enable auto score collection: virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy( training_id_eviction_trigger_count=260_000_000, # 260M training_id_keep_count=160_000_000, # 160M enable_auto_feature_score_collection=True, feature_score_mapping={ "sparse_public_original_content_creator": 1.0, }, feature_score_default_value=0.5, ), Additionally the counter collected previously during EC dedup is not used by kvzch backend, so this diff removed that counter and allow KJT to transfer a single float32 weight tensor to backend. This allows feature score collection for EBC since there could have another float weight for EBC pooling already. Reviewed By: EddyLXJ Differential Revision: D83945722
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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:
X-link: meta-pytorch/torchrec#3474
X-link: https://github.com/facebookresearch/FBGEMM/pull/2043
Enable feature score auto collection in ShardedEmbeddingCollection based on static feature to score mapping.
If user needs custom score for specific id, they can disable auto collection and then change model code explicitly to collect score for each id.
Here is the sample eviction policy config in embedding_table config to enable auto score collection:
virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy(
training_id_eviction_trigger_count=260_000_000, # 260M
training_id_keep_count=160_000_000, # 160M
enable_auto_feature_score_collection=True,
feature_score_mapping={
"sparse_public_original_content_creator": 1.0,
},
feature_score_default_value=0.5,
),
Differential Revision: D83945722