Question Answering on BBQ Overall Llama-3
80.7AccuracyCatRAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CatRAGModel=Llama-32026.03 | 80.7 | 32.3 | 0.01 | 97.6 | |
| Causal DebiasingModel=Llama-32026.03 | 78.6 | 28.2 | 0.1 | 84.1 | |
| SP DebiasingModel=Llama-32026.03 | 68.5 | 19.6 | 0.19 | 69.2 | |
| CE DebiasingModel=Llama-32026.03 | 64.2 | 15.3 | 0.28 | 54.3 | |
| Self DebiasingModel=Llama-32026.03 | 59.8 | 10.9 | 0.41 | 35.1 | |
| Base ModelModel=Llama-32026.03 | 48.9 | 0 | 0.63 | 0 |