HF detector
gravitee-small PII detection benchmark
Measured on 41 of 41 retained datasets under the same frozen scoring and masking protocol.
42.98%fully hidden annotations
50.49%detected / overlapped
5.36%extra masking
15,141CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Dates & times84.86%4,669 / 5,502
Phone numbers & email66.19%14,550 / 21,981
Bank accounts & cards55.74%2,928 / 5,253
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Logins & usernames4.33%493 / 11,382
Customer & employee IDs4.79%383 / 7,994
Other sensitive attributes15.68%503 / 3,207
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 17 | 71.73% | 77.55% | 5.08% |
| Russian | 19 | 28.60% | 37.50% | 6.12% |
| Multilingual | 5 | 59.45% | 64.60% | 2.75% |
Configuration, source and version
- Family
- HF
- Revision
- f8c27a85c51c0168f07b9dcf00265bf0a4097939
- Eligible datasets
- 41 / 41
- Character F1
- 0.496
- Untouched diagnostic
- 49.51% · 112,622 / 227,466
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- 0.660
- 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 ↓ |
|---|---|---|---|---|
| nerel | 24,369 | 2.76% | 3.77% | — |
| rubai-ru | 3,458 | 3.59% | 61.57% | — |
| factrueval | 7,966 | 3.72% | 4.07% | — |
| corrupt-secrets-issues | 288 | 11.11% | 27.08% | 3.94% |
| secrets-issues | 288 | 12.15% | 25.69% | 3.64% |
| alexen2 | 1,261 | 18.72% | 85.88% | 40.57% |
| jayguard | 1,195 | 22.01% | 29.62% | 2.95% |
| synth-secrets-ru | 581 | 23.41% | 42.51% | 10.98% |
| scanpatch | 8,708 | 24.82% | 35.94% | 5.96% |
| leaky-repo | 95 | 25.26% | 42.11% | 1.96% |
| leak-museum | 101 | 27.72% | 36.63% | 1.27% |
| synth-ru-tickets | 32,099 | 28.28% | 37.45% | — |
| multiconer-ru | 1,208 | 32.28% | 46.52% | 21.08% |
| nym-ru | 9,497 | 36.16% | 45.06% | — |
| synth-wiki-tables | 13,094 | 36.99% | 45.30% | — |
| creddata | 776 | 38.92% | 49.48% | 3.31% |
| dialogpii-multi | 10,345 | 39.35% | 46.01% | — |
| redact-ru | 9,230 | 39.88% | 48.16% | 4.28% |
| synth-env-configs | 392 | 40.56% | 76.79% | 10.81% |
| tab-echr | 3,830 | 40.94% | 56.48% | — |
| alrosait | 1,862 | 45.92% | 66.11% | 5.61% |
| synth-secrets-en | 581 | 46.30% | 70.91% | 11.10% |
| ameau01 | 2,242 | 46.65% | 47.73% | 1.81% |
| corrupt-redmadrobot | 5,531 | 52.99% | 60.40% | 4.23% |
| synth-jira-comments | 2,705 | 54.75% | 62.74% | — |
| secrets-rules | 746 | 58.18% | 72.92% | 25.36% |
| redmadrobot | 5,516 | 58.83% | 65.14% | 4.19% |
| redact-multi | 22,864 | 59.42% | 65.43% | — |
| nym-multi | 8,169 | 62.16% | 66.63% | 4.12% |
| corrupt-hivetrace | 1,667 | 62.57% | 74.93% | 11.52% |
| russian-pii-66k | 4,805 | 62.96% | 69.59% | — |
| dialogpii-en | 3,087 | 65.95% | 74.21% | 4.06% |
| hivetrace | 1,667 | 67.55% | 76.48% | 7.93% |
| arthur-passwords | 280 | 68.57% | 68.57% | 15.79% |
| nemotron-pii | 9,391 | 71.23% | 76.94% | 5.71% |
| kiji-multi | 7,565 | 72.12% | 73.96% | — |
| gretel-multi | 4,484 | 79.66% | 83.76% | 2.73% |
| tonicai | 2,417 | 82.25% | 86.22% | 3.70% |
| nym-en | 3,584 | 84.77% | 86.24% | 3.80% |
| kiji-en | 7,623 | 95.26% | 95.75% | — |
| privy | 1,899 | 99.21% | 99.58% | 1.91% |