List your company

Commonsense Reasoning on HellaSwag 5-shot

69.4AccuracyMoE 128e8a1s
55.98459.46762.9566.433May 22, 2026

Evaluation results

MethodLinks
MoE 128e8a1sModel=MoE 128e8a1s, Training Steps=100k2026.0569.4
MoE 128e8a4g1sModel=MoE 128e8a4g1s, Training Steps=100k2026.0569.3
DenseModel=Dense, Training Steps=100k2026.0556.5