List your company

Mathematical Reasoning on MATH (test) (Pass@1, RGR)

83.8Pass@1W2SR-P
30.13644.0685871.932May 26, 2025

Evaluation results

MethodLinks
W2SR-PStudent=Qwen2.5 32B, Teacher=Qwen2.5-14B (R), Training=Weak-to-Strong with Preference2025.0583.8200
W2SRStudent=Qwen2.5 32B, Teacher=Qwen2.5-14B (R), Training=Weak-to-Strong2025.0583.6187.5
W2SR-PStudent=Qwen2.5 32B, Teacher=Qwen2.5-7B (R), Training=Weak-to-Strong with Preference2025.0582.4104.55
GRPOStudent=Qwen2.5 32B, Teacher=None, Training=Reinforcement Learning2025.0582.2—
W2SRStudent=Qwen2.5 32B, Teacher=Qwen2.5-7B (R), Training=Weak-to-Strong2025.0581.686.36
W2SR-PStudent=Qwen2.5 14B, Teacher=Qwen2.5-7B (R), Training=Weak-to-Strong with Preference2025.0580.8107.14
GRPOStudent=Qwen2.5 14B, Teacher=None, Training=Reinforcement Learning2025.0580.6—
Qwen2.5-14B (R)Student=None, Teacher=Qwen2.5-14B (R), Training=Standalone Teacher2025.0580.6—
GRPOStudent=Qwen2.5 Math-7B, Teacher=None, Training=Reinforcement Learning2025.0580.2—
W2SRStudent=Qwen2.5 14B, Teacher=Qwen2.5-7B (R), Training=Weak-to-Strong2025.058078.57
W2SR-PStudent=Qwen2.5 Math-7B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong with Preference2025.057994.34
Qwen2.5-7B (R)Student=None, Teacher=Qwen2.5-7B (R), Training=Standalone Teacher2025.0577.8—
W2SR-PStudent=Qwen2.5 14B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong with Preference2025.057678.7
W2SR-PStudent=Qwen2.5 32B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong with Preference2025.057673.28
LIMOStudent=Qwen2.5 14B, Teacher=None, Training=SFT2025.0575.6—
LIMOStudent=Qwen2.5 32B, Teacher=None, Training=SFT2025.0575.4—
W2SRStudent=Qwen2.5 Math-7B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong2025.0571.257.55
LIMOStudent=Qwen2.5 Math-7B, Teacher=None, Training=SFT2025.0570.4—
W2SRStudent=Qwen2.5 14B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong2025.0570.251.85
W2SRStudent=Qwen2.5 32B, Teacher=Qwen2.5-1.5B (R), Training=Weak-to-Strong2025.0570.248.28
W2SR-PStudent=Qwen2.5 Math-7B, Teacher=Qwen2.5-0.5B (R), Training=Weak-to-Strong with Preference2025.0569.277.08
VanillaStudent=Qwen2.5 14B, Teacher=None, Training=Zero-shot standalone2025.0562.2—
VanillaStudent=Qwen2.5 32B, Teacher=None, Training=Zero-shot standalone2025.0560.4—
VanillaStudent=Qwen2.5 Math-7B, Teacher=None, Training=Zero-shot standalone2025.0560.2—
Qwen2.5-1.5B (R)Student=None, Teacher=Qwen2.5-1.5B (R), Training=Standalone Teacher2025.0559—
W2SRStudent=Qwen2.5 Math-7B, Teacher=Qwen2.5-0.5B (R), Training=Weak-to-Strong2025.055241.25
Qwen2.5-0.5B (R)Student=None, Teacher=Qwen2.5-0.5B (R), Training=Standalone Teacher2025.0532.2—