List your company

Mathematical Reasoning on AMC23 (pass@1, pass@256)

46.5Pass@1 ScoreGRPO + Branch
35.0638.034143.97May 15, 2026

Evaluation results

MethodLinks
GRPO + BranchBackbone=Qwen2.5-Math-7B-Base2026.0546.596.5
GRPOBackbone=Qwen2.5-Math-7B-Base2026.054396
GRPO + Forward KLBackbone=Qwen2.5-Math-7B-Base2026.054297.5
GRPO w/o KLBackbone=Qwen2.5-Math-7B-Base2026.054193
GRPO + RandomBackbone=Qwen2.5-Math-7B-Base2026.054196.5
GRPO + TokenBackbone=Qwen2.5-Math-7B-Base2026.0539.596
Base ModelBackbone=Qwen2.5-Math-7B-Base2026.0535.595