GLINER detector
gliner-pii-edge+cpu-int8 PII detection benchmark
Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.
9.58%fully hidden annotations
11.62%detected / overlapped
0.49%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Dates & times35.73%665 / 1,861
Documents & identifiers31.78%1,885 / 5,931
Bank accounts & cards28.83%542 / 1,880
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Network identifiers0.34%65 / 18,902
Organizations1.82%83 / 4,568
Logins & usernames5.23%460 / 8,791
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 4 | 11.69% | 12.89% | 0.52% |
| Russian | 4 | 7.68% | 10.10% | 0.42% |
| Multilingual | 1 | 17.48% | 18.77% | — |
Configuration, source and version
- Family
- GLINER
- Revision
- 9b7f39b0a2da971a5beea78d35f1539d4009c891
- Eligible datasets
- 9 / 41
- Character F1
- 0.149
- Untouched diagnostic
- 88.38% · 63,718 / 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 ↓ |
|---|---|---|---|---|
| synth-ru-tickets | 32,099 | 1.24% | 2.76% | — |
| tab-echr | 3,830 | 3.29% | 3.81% | — |
| ameau01 | 2,242 | 5.93% | 6.42% | 0.38% |
| secrets-issues | 288 | 11.11% | 15.63% | 0.55% |
| nemotron-pii | 9,391 | 16.51% | 18.05% | 0.00% |
| kiji-multi | 7,565 | 17.48% | 18.77% | — |
| nym-ru | 9,497 | 18.26% | 20.96% | — |
| redmadrobot | 5,516 | 19.07% | 23.95% | 0.42% |
| hivetrace | 1,667 | 33.53% | 43.73% | 0.43% |