Faithfulness Evaluation on Alpaca 800 samples
62.1BLEURandom
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RandomModel=Llama-2 (7B-Chat)2024.05 | 62.1 | 53.9 | 54 | 53.5 | 82.4 | 87.1 | 0.104 | |
| AttentionModel=Llama-2 (7B-Chat)2024.05 | 54.6 | 46.2 | 46.4 | 45.8 | 71.3 | 80.2 | 0.168 | |
| Last-AttentionModel=Llama-2 (7B-Chat)2024.05 | 54 | 44.7 | 45.3 | 44.4 | 71.3 | 78.7 | 0.191 | |
| Integrated-GradientModel=Llama-2 (7B-Chat)2024.05 | 49.7 | 39.5 | 39.6 | 39 | 68.7 | 70.4 | 0.275 | |
| JoPAModel=Llama-2 (7B-Chat)2024.05 | 47.9 | 38.3 | 37.6 | 37.2 | 64.2 | 56.5 | 0.479 |