Lab 04: Dataset Contamination Checker
Series: AI Eval Labs · Phase: Foundations · Depends on: None
Core scope
- [ ] N-gram overlap scan between eval set and a public training corpus sample
- [ ] Embedding similarity scan for near-duplicates
- [ ] Contamination report with thresholds
Stretch
- [ ] MinHash/LSH to scale to large corpora
Links
- Code:
series/ai-eval/04-contamination-checker - Demo: coming soon
- Write-up: coming soon on markrox.ai