filler_gap_pp on CausalGym
97.64CompletenessFGSM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FGSMModel=Meta-Llama-3-8B-Instruct2026.05 | 97.64 | 5.01 | 9.54 | |
| HDMIModel=Meta-Llama-3-8B-Instruct2026.05 | 97.1 | 44.12 | 60.67 | |
| PGDModel=Meta-Llama-3-8B-Instruct2026.05 | 84.8 | 22.9 | 36.06 | |
| AlterRepModel=Meta-Llama-3-8B-Instruct2026.05 | 80.95 | 73.89 | 77.26 | |
| HDMIModel=EleutherAI/pythia-70m2026.05 | 74.04 | 62.75 | 67.93 | |
| AlterRepModel=EleutherAI/pythia-70m2026.05 | 51 | 45.82 | 47.82 | |
| PGDModel=EleutherAI/pythia-70m2026.05 | 45.61 | 93.1 | 61.22 | |
| FGSMModel=EleutherAI/pythia-70m2026.05 | 44.53 | 99.51 | 61.53 |