Question Answering on TruthfulQA (Accuracy and Uncertainty Metrics)
0.824AccUDPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| UDPO# compromised agents=12026.02 | 0.824 | 0.108 | 0.076 | 0.118 | |
| UDPO# compromised agents=22026.02 | 0.718 | 0.165 | 0.118 | 0.182 | |
| UDPO# compromised agents=32026.02 | 0.707 | 0.242 | 0.185 | 0.255 | |
| RMAAC# compromised agents=12026.02 | 0.661 | 0.212 | 0.165 | 0.285 | |
| MAPPO# compromised agents=12026.02 | 0.635 | 0.225 | 0.178 | 0.305 | |
| Standard MAD# compromised agents=12026.02 | 0.587 | 0.247 | 0.212 | 0.362 | |
| RMAAC# compromised agents=22026.02 | 0.563 | 0.275 | 0.228 | 0.451 | |
| RMAAC# compromised agents=32026.02 | 0.531 | 0.342 | 0.295 | 0.47 | |
| MAPPO# compromised agents=22026.02 | 0.524 | 0.298 | 0.245 | 0.418 | |
| MAPPO# compromised agents=32026.02 | 0.512 | 0.372 | 0.315 | 0.528 | |
| Standard MAD# compromised agents=22026.02 | 0.459 | 0.385 | 0.298 | 0.434 | |
| Standard MAD# compromised agents=32026.02 | 0.238 | 0.593 | 0.582 | 0.576 |