List your company

Natural Language Inference on HANS evaluated subset

157Rule-bearingMechaRule (Qwen2-1.5B)
10.3648.4386.5124.57May 4, 2026

Evaluation results

MethodLinks
MechaRule (Qwen2-1.5B)Condition=exhaustive singletons2026.051570.91111432.214.2
MechaRule (Qwen2-7B)Condition=spectral only2026.05330.9242461.219
MechaRule (Qwen2-1.5B)Condition=spectral only2026.05290.9452532.213.2
MechaRule (Qwen2-7B)Condition=rule + spectral2026.05290.9252161.518.6
MechaRule (Qwen2-7B)Condition=rule + random2026.05280.9152161.218.8
MechaRule (Qwen2-1.5B)Condition=rule + spectral2026.05260.9342432.213.4
MechaRule (Qwen2-7B)Condition=fake rule + spectral2026.05260.8991761.517.6
MechaRule (Qwen2-1.5B)Condition=rule + random2026.05240.9452232.213.5
MechaRule (Qwen2-1.5B)Condition=fake rule + spectral2026.05160.931432.211.5