HF detector
ner-ru-gherman+cpu-int8 PII detection benchmark
Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.
17.49%fully hidden annotations
32.56%detected / overlapped
0.11%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
People's names62.90%6,943 / 11,039
Addresses & locations51.24%4,564 / 8,907
Logins & usernames11.39%1,001 / 8,791
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Other sensitive attributes0.00%0 / 224
Customer & employee IDs0.00%0 / 1,819
Network identifiers0.00%0 / 18,902
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 4 | 22.98% | 39.51% | 0.10% |
| Russian | 4 | 10.36% | 26.61% | 0.16% |
| Multilingual | 1 | 52.07% | 56.38% | — |
Configuration, source and version
- Family
- HF
- Revision
- fc6b2c5a2c5d7c82da7416f7fd9c055159bbb984
- Eligible datasets
- 9 / 41
- Character F1
- 0.322
- Untouched diagnostic
- 67.44% · 48,624 / 72,095
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- Not measured
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| secrets-issues | 288 | 0.00% | 0.00% | 0.10% |
| hivetrace | 1,667 | 0.84% | 29.15% | 0.08% |
| synth-ru-tickets | 32,099 | 2.65% | 20.80% | — |
| ameau01 | 2,242 | 9.23% | 27.88% | 0.05% |
| tab-echr | 3,830 | 13.81% | 55.56% | — |
| nym-ru | 9,497 | 24.77% | 36.38% | — |
| nemotron-pii | 9,391 | 30.70% | 36.96% | 0.00% |
| redmadrobot | 5,516 | 33.30% | 42.84% | 0.18% |
| kiji-multi | 7,565 | 52.07% | 56.38% | — |