Skip to content

fix(embedding): reject non-progressing chunk windows - #142

Open
deepujain wants to merge 3 commits into
NVIDIA:mainfrom
deepujain:fix/139-non-progressing-embedding-windows
Open

deepujain wants to merge 3 commits into
NVIDIA:mainfrom
deepujain:fix/139-non-progressing-embedding-windows

Conversation

@deepujain

Copy link
Copy Markdown
Contributor

Summary

  • reject zero-sized embedding chunk windows
  • reject negative overlap and overlap greater than or equal to the chunk size
  • fail before entering the fixed-window splitter or contacting the embedding provider
  • add focused regression coverage for zero and non-progressing windows

Closes #139.

Verification

  • .venv/bin/python -m pytest -q tests/embedding/test_client.py — 31 passed
  • make PYTHON=.venv/bin/python lint — passed
  • .venv/bin/python -m build — sdist and wheel built successfully
  • git diff --check origin/main...HEAD — passed
  • make PYTHON=.venv/bin/python test — 6,054 passed, 18 skipped, 4 deselected, 4 unrelated baseline failures: one CLI golden-surface drift and three Anthropic SDK httpx2 compatibility expectations

Release Impact

Embedding chunk configuration now fails with an actionable SimilarityConfigError instead of allowing a zero or negative step to stall the fixed-size splitter. The change is backward compatible for valid chunk settings.

Signed-off-by: Deepak Jain <deepujain@gmail.com>
@deepujain
deepujain force-pushed the fix/139-non-progressing-embedding-windows branch from b4d3486 to e18c1f6 Compare September 15, 2026 00:18
Comment thread src/skillevaluator/embedding/client.py
Signed-off-by: Deepak Jain <deepujain@gmail.com>
@rng1995

rng1995 commented Sep 16, 2026

Copy link
Copy Markdown
Collaborator

@deepujain Please resolve merge conflicts

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.

Reject non-progressing embedding chunk windows before entering the splitter

2 participants