List your company

Mathematical Reasoning on GSM8K full (test) (Accuracy, Avg.)

79.8AccuracySAE (Global)
11.05628.90346.7564.597Feb 9, 2026

Evaluation results

MethodLinks
SAE (Global)Fusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct, scope=global2026.0279.848.4
PSOFusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct2026.0278.0147.2
SAE (Local)Fusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct, scope=local2026.0277.4847.8
Weight AverageFusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct2026.0274.246.4
Task ArithmeticFusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct2026.0274.146.4
RankmeanFusion Setting=Math + Multilingual, Backbone=LLaMA-3.2-3B-Instruct2026.0213.715.7