HF detector
davlan-xlmr+cpu-int8 PII detection benchmark
Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.
18.67%fully hidden annotations
25.45%detected / overlapped
0.26%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
People's names69.12%7,630 / 11,039
Addresses & locations47.78%4,256 / 8,907
Organizations33.30%1,521 / 4,568
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 | 29.50% | 42.63% | 0.26% |
| Russian | 4 | 10.09% | 15.45% | 0.24% |
| Multilingual | 1 | 51.45% | 54.21% | — |
Configuration, source and version
- Family
- HF
- Revision
- 253f557bd8249b8515114cfd7f71974fe5fa4d2f
- Eligible datasets
- 9 / 41
- Character F1
- 0.287
- Untouched diagnostic
- 74.55% · 53,745 / 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% | 0.35% | 0.11% |
| synth-ru-tickets | 32,099 | 4.52% | 9.14% | — |
| hivetrace | 1,667 | 12.42% | 21.30% | 0.10% |
| nym-ru | 9,497 | 17.89% | 25.93% | — |
| ameau01 | 2,242 | 18.73% | 20.96% | 1.02% |
| redmadrobot | 5,516 | 28.41% | 32.31% | 0.27% |
| nemotron-pii | 9,391 | 30.24% | 34.41% | 0.00% |
| tab-echr | 3,830 | 36.21% | 78.64% | — |
| kiji-multi | 7,565 | 51.45% | 54.21% | — |