List your company

Physical Commonsense Reasoning on PIQA (5-shot)

77.8AccuracyMoE 128e8a1s
72.39273.79675.276.604May 22, 2026

Evaluation results

MethodLinks
MoE 128e8a1sModel=MoE 128e8a1s, Training Steps=100k2026.0577.8
MoE 128e8a4g1sModel=MoE 128e8a4g1s, Training Steps=100k2026.0577.7
DenseModel=Dense, Training Steps=100k2026.0572.6