List your company

Mathematical Reasoning on MNNS (val)

98.94AccuracyCoT2
67.126475.385783.64591.9043May 29, 2025

Evaluation results

MethodLinks
CoT2Backbone=2-layer, 2-head GPT2 (d=32), Budget=Full, Input digits=4 digits in 1–92025.0598.94
COCONUTBackbone=2-layer, 2-head GPT2 (d=32), Input digits=4 digits in 1–92025.0592.58
Discrete CoTBackbone=2-layer, 2-head GPT2 (d=32), Budget=B=1, Input digits=4 digits in 1–92025.0584.92
Discrete no-CoTBackbone=2-layer, 2-head GPT2 (d=32), Input digits=4 digits in 1–92025.0568.35