List your company

Mathematical Reasoning on GSM8K (Parameter & Training Efficiency)

509.7Parameters (Full) (M)Hot-25% (MoE-Sieve)
303.572357.086410.6464.114Mar 25, 2026

Evaluation results

MethodLinks
Hot-25% (MoE-Sieve)Backbone Model=Qwen2026.03509.7151.370.32.04606713149
Hot-25% (MoE-Sieve)Backbone Model=OLMoE2026.03311.58572.71.2534073.415450