Data Contamination Detection on DETCON Code Generation
71.5AccuracyCDD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CDD2024.02 | 71.5 | 69.4 | 76.1 | 72.4 | |
| Min-k% Prob2024.02 | 56.3 | 52.4 | 56.5 | 55 | |
| N-gramlevel=token-level2024.02 | 54.1 | 30.2 | — | 42.2 | |
| LLM Decontaminatornote=needs additional conditions2024.02 | 53.5 | 57.8 | — | 55.6 | |
| Embedding similarity2024.02 | 52.4 | 56.9 | 57.1 | 55.4 | |
| Perplexity2024.02 | 51.3 | 59.3 | 49.1 | 53.2 | |
| N-gramlevel=char-level2024.02 | 48.4 | 59.3 | — | 53.8 |