Add batched pairwise similarity method for Semantic Dedup #581
+160
−24
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
Signed-off-by: Praateek <[email protected]>
Signed-off-by: Praateek <[email protected]>
praateekmahajan
requested review from
sarahyurick,
VibhuJawa,
ryantwolf and
ayushdg
Signed-off-by: Praateek <[email protected]>
Signed-off-by: Praateek <[email protected]>
sarahyurick
reviewed
Mar 7, 2025
nemo_curator/modules/semantic_dedup/semanticclusterleveldedup.py
Outdated
Show resolved
Hide resolved
nemo_curator/modules/semantic_dedup/semanticclusterleveldedup.py
Outdated
Show resolved
Hide resolved
nemo_curator/utils/semdedup_utils.py
Outdated
Comment on lines
194
to
195
| # Compute pairwise cosine similarity | ||
| pairwise_sim_matrix = cluster_reps @ (cluster_reps.T) |
There was a problem hiding this comment.
Haha I don't think I've used @ before. Would there be any advantage to using torch.mm, torch.matmul, etc.?
| @@ -681,6 +681,7 @@ | |||
| " id_column_type=\"str\",\n", | |||
| " embedding_col=\"image_embedding\",\n", | |||
| " which_to_keep=\"hard\",\n", | |||
| " batched_cosine_similarity=1024,\n", | |||
There was a problem hiding this comment.
Nope, do we always manually run these notebooks for such PRs? That'll be a time sink but I'm okay to do it that's the practice
There was a problem hiding this comment.
If it is expected to produce the same results as before, it is okay with me. Sometimes I leave notebooks unchanged (or add changes to have it keep the previous default) if the output is expected to change, so that the user won't be confused when their cell outputs are different than the ones on GitHub.
There was a problem hiding this comment.
But it sounds like that isn't the case here?
praateekmahajan
changed the title
Signed-off-by: Praateek <[email protected]>
Signed-off-by: Praateek <[email protected]>
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.
Description
Resolves #520
Currently if a single cluster is large enough, we'll likely OOM since
M @ M.TrequiresN**2storage. A batched version breaks it up into smaller batches B and performsM @ B.T. The only thing we need to be careful is how to zero out the diagonals and get the upper triangular matrix.other nits
M @ M.Tresults in an absolute max value of 1_semdeduptopairwise_similarityUsage
# Add snippet demonstrating usageChecklist