HF detector
davlan-mbert+cpu-int8 PII detection benchmark
Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.
24.56%fully hidden annotations
30.43%detected / overlapped
0.76%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
People's names87.64%9,675 / 11,039
Organizations60.99%2,786 / 4,568
Addresses & locations56.90%5,068 / 8,907
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Other sensitive attributes0.00%0 / 224
Customer & employee IDs0.00%0 / 1,819
Dates & times0.00%0 / 1,861
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 4 | 41.53% | 52.36% | 0.91% |
| Russian | 4 | 14.31% | 19.31% | 0.37% |
| Multilingual | 1 | 55.31% | 56.44% | — |
Configuration, source and version
- Family
- HF
- Revision
- e756de7f7b8f64fea0c3d7c3872f1322fab747b1
- Eligible datasets
- 9 / 41
- Character F1
- 0.350
- Untouched diagnostic
- 69.57% · 50,159 / 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% | 1.39% | 0.53% |
| synth-ru-tickets | 32,099 | 7.25% | 11.32% | — |
| hivetrace | 1,667 | 13.56% | 23.46% | 0.21% |
| nym-ru | 9,497 | 25.63% | 34.04% | — |
| ameau01 | 2,242 | 26.45% | 30.51% | 2.87% |
| redmadrobot | 5,516 | 36.11% | 39.16% | 0.40% |
| nemotron-pii | 9,391 | 41.11% | 44.96% | 0.00% |
| tab-echr | 3,830 | 54.49% | 87.15% | — |
| kiji-multi | 7,565 | 55.31% | 56.44% | — |