List your company

Mathematical Reasoning on GSM8K (Accuracy, Performance Gain)

96.81Accuracyl2-guided Selection
88.001290.288192.57594.8619Jun 4, 2026

Evaluation results

MethodLinks
l2-guided SelectionModel=Qwen3-32B2026.0696.810.38
Base (Mean@8)Model=Qwen3-32B2026.0696.44—
l2-guided SelectionModel=Qwen3-14B2026.0696.360.24
l2-guided SelectionModel=Qwen3-4B2026.0696.20.83
Base (Mean@8)Model=Qwen3-14B2026.0696.13—
l2-guided SelectionModel=Qwen3-8B2026.0696.060.16
Base (Mean@8)Model=Qwen3-8B2026.0695.91—
Base (Mean@8)Model=Qwen3-4B2026.0695.41—
l2-guided SelectionModel=LLaMA-70B2026.0695.140.88
Base (Mean@8)Model=LLaMA-70B2026.0694.31—
l2-guided SelectionModel=Qwen3-1.7B2026.0691.211.31
l2-guided SelectionModel=LLaMA-8B2026.0690.822.81
Base (Mean@8)Model=Qwen3-1.7B2026.0690.03—
Base (Mean@8)Model=LLaMA-8B2026.0688.34—