Data Contamination Detection on DETCON Logical Reasoning
70.6AccuracyCDD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CDD2024.02 | 70.6 | 76.5 | 84.6 | 77.3 | |
| N-gramlevel=token-level2024.02 | 65.6 | 49.8 | — | 57.7 | |
| Embedding similarity2024.02 | 59.2 | 64.5 | 66.8 | 63.5 | |
| N-gramlevel=char-level2024.02 | 56.4 | 67 | — | 61.7 | |
| Min-k% Prob2024.02 | 52.7 | 67.7 | 69.8 | 63.4 | |
| LLM Decontaminatornote=needs additional conditions2024.02 | 50.9 | 43.3 | — | 47.1 | |
| Perplexity2024.02 | 49.7 | 66.4 | 69.9 | 62 |