PII Span Detection (Span-Correct) on AI4Privacy (test)
99.4AccuracyDeepSeek-Q1
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-Q1Training Protocol=Instruction-Tuned (Inst.), Evaluation Source=PRvL Baseline2025.12 | 99.4 | 97.3 | 98.1 | |
| Llama3.1-8BTraining Protocol=Fine-Tuned, Evaluation Source=PRvL Baseline2025.12 | 98.6 | 91.5 | 96.9 | |
| Mistral_2Architecture=Mistral 7B, Evaluation Protocol=Relaxed Metrics2025.12 | 98.5 | 96.3 | 91.8 | |
| T5Checkpoint Selection=Best Checkpoint, Evaluation Protocol=Relaxed Metrics2025.12 | 97.2 | 89.1 | 91 | |
| Mistral_1Architecture=Mistral 7B, Evaluation Protocol=Relaxed Metrics2025.12 | 93.6 | 64.7 | 92.1 | |
| T5Training Protocol=Fine-Tuned, Evaluation Source=PRvL Baseline2025.12 | 88.3 | 72.7 | 83 |