Filler-gap dependency hierarchy on CausalGym filler_gap_hierarchy
100CompletenessHDMI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HDMIModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 84.14 | 91.38 | |
| FGSMModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 100 | 100 | |
| HDMIModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| FGSMModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| PGDModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| HDMIModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 84.14 | 91.38 | |
| HDMIModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| FGSMModel=Meta-Llama-3-8B-Instruct2026.05 | 100 | 100 | 100 | |
| FGSMModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| PGDModel=EleutherAI/pythia-70m2026.05 | 100 | 100 | 100 | |
| AlterRepModel=Meta-Llama-3-8B-Instruct2026.05 | 99 | 84.43 | 91.14 | |
| AlterRepModel=Meta-Llama-3-8B-Instruct2026.05 | 99 | 84.43 | 91.14 | |
| PGDModel=Meta-Llama-3-8B-Instruct2026.05 | 83 | 100 | 90.71 | |
| PGDModel=Meta-Llama-3-8B-Instruct2026.05 | 83 | 100 | 90.71 | |
| AlterRepModel=EleutherAI/pythia-70m2026.05 | 82.99 | 100 | 90.7 | |
| AlterRepModel=EleutherAI/pythia-70m2026.05 | 82.99 | 100 | 90.7 |