List your company
ResearchFormerly WizWand

Structural Erasure on MUSE

1.979CADSimNPO
-0.079160.455170.98951.52383May 14, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SimNPOModel=Llama-3.1-8B-Instruct2026.051.9791.9791.104
MANSUModel=Llama-3.1-8B-Instruct2026.051.6710.3180.097
Global GAModel=Llama-3.1-8B-Instruct2026.051.661.661.187
NPOModel=Llama-3.1-8B-Instruct2026.051.6351.6351.15
Surgical GAModel=Llama-3.1-8B-Instruct2026.051.5991.5990.469
GU+SimNPOModel=Llama-3.1-8B-Instruct2026.051.5221.5221.256
SimNPOModel=Qwen-3-8B2026.051.2991.2990.828
NPOModel=Qwen-3-8B2026.051.231.230.77
MANSUModel=Qwen-3-8B2026.051.1430.4410.182
GU+SimNPOModel=Qwen-3-8B2026.051.1121.1120.757
Global GAModel=Qwen-3-8B2026.051.1061.1060.721
Surgical GAModel=Qwen-3-8B2026.051.091.090.568
LUNARModel=Llama-3.1-8B-Instruct2026.050.0451.2480.328
LUNARModel=Qwen-3-8B2026.050.0431.2010.316
Zero-shotModel=Llama-3.1-8B-Instruct2026.05000
Zero-shotModel=Qwen-3-8B2026.05000