GLINER2 detector
gliner2-vladlinv+homoglyph PII detection benchmark
Measured on 3 of 41 retained datasets under the same frozen scoring and masking protocol.
45.00%fully hidden annotations
46.98%detected / overlapped
0.07%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Phone numbers & email75.61%589 / 779
People's names57.33%853 / 1,488
Documents & identifiers53.30%1,431 / 2,685
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Passwords, keys & tokens7.57%29 / 383
Network identifiers9.14%33 / 361
Addresses & locations18.62%264 / 1,418
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 1 | 4.17% | 9.72% | 0.04% |
| Russian | 2 | 46.64% | 48.47% | 0.15% |
Configuration, source and version
- Family
- GLINER2
- Revision
- d4c05667b10d8e8db0f0626934135731d0530983
- Eligible datasets
- 3 / 41
- Character F1
- 0.471
- Untouched diagnostic
- 53.02% · 3,969 / 7,486
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 0.162
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| corrupt-secrets-issues | 288 | 4.17% | 9.72% | 0.04% |
| corrupt-redmadrobot | 5,531 | 42.38% | 43.34% | 0.10% |
| corrupt-hivetrace | 1,667 | 60.77% | 65.51% | 0.44% |