Skip to content

feat(semdedup): bound pairwise precision and memory - #2318

Draft
praateekmahajan wants to merge 1 commit into
NVIDIA-NeMo:mainfrom
praateekmahajan:praateekm/nmcur-318-semdedup-pairwise-precision
Draft

feat(semdedup): bound pairwise precision and memory#2318
praateekmahajan wants to merge 1 commit into
NVIDIA-NeMo:mainfrom
praateekmahajan:praateekm/nmcur-318-semdedup-pairwise-precision

Conversation

@praateekmahajan

Copy link
Copy Markdown
Contributor

Description

Bound Pairwise semantic-dedup memory and add configurable FP16/FP32 compute precision.

Linear: NMCUR-318

Validation

  • Focused GPU suite and FP16/FP32 benchmark smoke covered
  • Post-rebase container verification is running

@copy-pr-bot

copy-pr-bot Bot commented Aug 17, 2026

Copy link
Copy Markdown

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

@github-actions

Copy link
Copy Markdown
Contributor

Comment thread benchmarking/scripts/semdedup_identification_benchmark.py Outdated
Comment thread benchmarking/scripts/semdedup_pairwise_benchmark.py Outdated
Comment thread benchmarking/scripts/semdedup_identification_benchmark.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread tests/benchmarking/test_semdedup_identification_benchmark.py Outdated
Comment thread tests/stages/deduplication/semantic/test_pairwise.py Outdated
Comment thread tests/stages/text/deduplication/test_semantic.py Outdated
assert stages[1].profile is True
assert stages[1].verbose is True

def test_new_options_preserve_existing_positional_order(self) -> None:

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this test is very local, and against what AGENTS.md recommends

Comment thread tests/stages/deduplication/semantic/test_workflow.py Outdated
Comment thread tests/stages/deduplication/semantic/test_workflow.py Outdated
Comment thread tests/stages/text/deduplication/test_semantic.py Outdated
Comment thread tests/stages/deduplication/semantic/test_pairwise.py Outdated
Comment thread tests/stages/deduplication/semantic/test_pairwise.py Outdated
@praateekmahajan
praateekmahajan force-pushed the praateekm/nmcur-318-semdedup-pairwise-precision branch from 44c3013 to 5586523 Compare August 18, 2026 00:30
return None


def release_cached_gpu_memory() -> None:

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ayushdg can you review this, to see if it's not sloppy? I'm wondering if we can then also add this in the minhash teardown part

Comment thread tests/stages/deduplication/semantic/test_pairwise.py
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread nemo_curator/stages/deduplication/semantic/pairwise.py Outdated
Comment thread tests/stages/deduplication/semantic/test_pairwise.py
Comment thread nemo_curator/stages/deduplication/gpu_utils.py Outdated
@praateekmahajan
praateekmahajan force-pushed the praateekm/nmcur-318-semdedup-pairwise-precision branch 2 times, most recently from d63a7f3 to cd18cc1 Compare August 21, 2026 01:31
@praateekmahajan

Copy link
Copy Markdown
Contributor Author

/ok to test cd18cc1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant