HF detector
ru-legal-ner+ov100 PII detection benchmark
Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.
38.48%fully hidden annotations
60.80%detected / overlapped
9.95%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
People's names71.30%17,362 / 24,350
Dates & times61.40%369 / 601
Phone numbers & email59.11%3,397 / 5,747
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Network identifiers2.15%376 / 17,498
Other sensitive attributes14.51%129 / 889
Organizations24.92%3,270 / 13,123
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 6 | 21.00% | 37.99% | 10.78% |
| Russian | 5 | 42.55% | 66.12% | 4.48% |
Configuration, source and version
- Family
- HF
- Revision
- c0313a42ac147ccc38f5c3b0b3e77cab53208683
- Eligible datasets
- 11 / 41
- Character F1
- 0.479
- Untouched diagnostic
- 39.20% · 36,198 / 92,332
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 0.129
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| tab-echr | 3,830 | 6.32% | 19.19% | — |
| dialogpii-en | 3,087 | 7.52% | 16.33% | 0.75% |
| leak-museum | 101 | 25.74% | 48.51% | 5.86% |
| nemotron-pii | 9,391 | 26.39% | 46.87% | 21.32% |
| synth-ru-tickets | 32,099 | 29.24% | 72.11% | — |
| nerel | 24,369 | 49.39% | 56.19% | — |
| factrueval | 7,966 | 53.13% | 55.99% | — |
| secrets-issues | 288 | 56.60% | 86.81% | 10.23% |
| jayguard | 1,195 | 59.41% | 78.33% | 4.17% |
| redact-ru | 9,230 | 59.51% | 78.67% | 16.71% |
| creddata | 776 | 68.17% | 89.95% | 14.02% |