agr_sv_num_pp on CausalGym
100CompletenessHDMI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HDMIModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 93.62 | 96.71 | |
| FGSMModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 83.14 | 90.8 | |
| PGDModel=EleutherAI/pythia-70m2026.05 | 100 | 85.5 | 92.18 | |
| HDMIModel=EleutherAI/pythia-70m2026.05 | 99.84 | 94.09 | 96.88 | |
| FGSMModel=EleutherAI/pythia-70m2026.05 | 99.84 | 49.06 | 65.79 | |
| AlterRepModel=Meta-Llama-3-8B-Instruct2026.05 | 95 | 86.8 | 90.72 | |
| AlterRepModel=EleutherAI/pythia-70m2026.05 | 50 | 51.08 | 50.53 | |
| PGDModel=Meta-Llama-3-8B-Instruct2026.05 | 36.5 | 36.48 | 36.49 |