[BUG] l2_exp & KL distances got randomly round-off errors in half-float32 mixed precision on self-neighboring
Open
Nobody has claimed this yet.
bug
- Dominant language
- Cuda
- Stars
- 854
- Forks
- 236
- Avg merge
- 3d 3h
- Merged PRs (30d)
- 62
Description
- This PR resolved this issue partially
- Considering the issue becomes more serious as the
klarger, there's a need to discuss a perfect solution- It's hard to determine the thresholds empirically for
get_clamp_precision. Please refer to the code: https://github.com/rapidsai/cuvs/blob/branch-25.02/cpp/src/distance/detail/distance_ops/l2_exp.cuh#L32
- It's hard to determine the thresholds empirically for
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Review the partial resolution in PR 596 and inspect cpp/src/distance/detail/distance_ops/l2_exp.cuh, especially get_clamp_precision. Investigate why self-neighboring l2_exp and KL distances still show round-off errors in half-float32 mixed precision as k increases. Done means establishing and validating a robust solution rather than relying on empirically chosen thresholds.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- performance, search
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100