List your company

Circuit Recovery on Aggregate IOI, MCQA, Arithmetic, ARC

87CPR Faithfulness ScoreNAP-IG-Inputs
23.5640.0356.572.97Apr 27, 2026

Evaluation results

MethodLinks
NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=Gold (Upper Bound)2026.0487———————
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=Gold (Upper Bound)2026.0468———————
NAPMethod Category=NAP, Transfer Setting=Gold (Upper Bound)2026.0466———————
NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=DFA (Far/Zero-Shot)2026.0431———————
NAPMethod Category=NAP, Transfer Setting=DFA (Far/Zero-Shot)2026.0431———————
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (Far/Zero-Shot)2026.0430———————
NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=DFA (In-Distribution)2026.0429———————
NAP-IG-InputsMethod Category=NAP-IG-Inputs, Transfer Setting=DFA (Near-Distribution)2026.0429———————
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (In-Distribution)2026.0428———————
NAPMethod Category=NAP, Transfer Setting=DFA (In-Distribution)2026.0427———————
NAPMethod Category=NAP, Transfer Setting=DFA (Near-Distribution)2026.0427———————
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (Near-Distribution)2026.0427———————
BaselineMethod Category=Baseline, Transfer Setting=Random Alignment2026.0426———————
NAPMethod Category=NAP, Transfer Setting=Gold (Upper Bound), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.250.350.250.250.370.340.3
NAPMethod Category=NAP, Transfer Setting=DFA (In-Distribution), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.320.490.260.260.460.520.39
NAPMethod Category=NAP, Transfer Setting=DFA (Near-Distribution), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04———0.390.250.440.480.39
NAPMethod Category=NAP, Transfer Setting=DFA (Far/Zero-Shot), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.270.460.280.260.440.480.37
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=Gold (Upper Bound), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.380.350.250.250.430.350.33
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (In-Distribution), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.330.390.530.450.410.440.42
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (Near-Distribution), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04———0.720.280.410.370.44
NAP-IG-ActivationsMethod Category=NAP-IG-Activations, Transfer Setting=DFA (Far/Zero-Shot), Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.330.460.820.280.340.510.46
Random AlignmentMethod Category=Baseline, Transfer Setting=Random Alignment, Source Model=Llama-3 (1B), Target Model=Llama-3 (3B)2026.04—0.250.250.250.250.250.250.25