Question Answering on CSQA (test) (Acc, F1, Ref, ASR)
87.2AccuracyML Routing
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ML RoutingAdversarial attack=TextAttack2026.06 | 87.2 | 86.96 | 86.47 | 14.47 | |
| LWV FusionAdversarial attack=TextAttack2026.06 | 87.02 | 86.77 | 83.61 | 17.47 | |
| Geometric-Based ModelAdversarial attack=TextAttack2026.06 | 70.54 | 67.28 | — | 27.53 | |
| Uncertainty-Based ModelAdversarial attack=TextAttack2026.06 | 68.6 | 68.7 | 34.33 | 45.67 | |
| Entropy-Based ModelAdversarial attack=TextAttack2026.06 | 62.9 | 55.12 | 36.88 | 51.97 | |
| Base Model (no defence)Adversarial attack=TextAttack2026.06 | 48.03 | 48.03 | — | 63.12 | |
| FreeLBAdversarial attack=TextAttack2026.06 | 22.11 | 22.11 | — | 77.89 |