Circuit Recovery on MCQA
1.13CPR (Faithfulness Score)NAP-IG-Inputs
Evaluation Results
| Method | Links | |
|---|---|---|
| NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=Gold (Upper Bound)2026.04 | 1.13 | |
| NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=Gold (Upper Bound)2026.04 | 0.9 | |
| NAPMethod Category=NAP, Transfer Setting=Gold (Upper Bound)2026.04 | 0.82 | |
| NAP-IG-InputsTransfer Setting=DFA (Far/Zero-Shot), Model Pair=Llama-3 (1B -> 3B)2026.04 | 0.48 | |
| NAP-IG-InputsTransfer Setting=Gold (Upper Bound), Model Pair=Llama-3 (1B -> 3B)2026.04 | 0.47 | |
| NAP-IG-InputsTransfer Setting=DFA (In-Distribution), Model Pair=Llama-3 (1B -> 3B)2026.04 | 0.41 | |
| NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=DFA (Far/Zero-Shot)2026.04 | 0.33 | |
| NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=DFA (In-Distribution)2026.04 | 0.32 | |
| NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (Far/Zero-Shot)2026.04 | 0.3 | |
| BaselineMethod Category=Baseline, Transfer Setting=Random Alignment2026.04 | 0.28 | |
| NAPMethod Category=NAP, Transfer Setting=DFA (Far/Zero-Shot)2026.04 | 0.28 | |
| NAPMethod Category=NAP, Transfer Setting=DFA (In-Distribution)2026.04 | 0.26 | |
| NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (In-Distribution)2026.04 | 0.25 | |
| BaselineTransfer Setting=Random Alignment, Model Pair=Llama-3 (1B -> 3B)2026.04 | 0.25 |