agr_refl_num_pp on CausalGym
100CompletenessHDMI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HDMIModel=EleutherAI/pythia-70m2026.05 | 100 | 89.83 | 94.64 | |
| HDMIModel=Meta-Llama-3-8B-Instruct2026.05 | 99.02 | 82.72 | 90.14 | |
| AlterRepModel=Meta-Llama-3-8B-Instruct2026.05 | 99 | 65.4 | 78.77 | |
| FGSMModel=EleutherAI/pythia-70m2026.05 | 96.42 | 92.06 | 94.19 | |
| PGDModel=EleutherAI/pythia-70m2026.05 | 61.42 | 100 | 76.1 | |
| AlterRepModel=EleutherAI/pythia-70m2026.05 | 54.49 | 88.43 | 67.43 | |
| FGSMModel=Meta-Llama-3-8B-Instruct2026.05 | 53.41 | 98.81 | 69.34 | |
| PGDModel=Meta-Llama-3-8B-Instruct2026.05 | 33.35 | 51.55 | 40.5 |