List your company

Math Reasoning on Minerva (Accuracy)

40.8AccuracySFT
6.06415.08224.133.118Jul 2, 2025

Evaluation results

MethodLinks
SFTBase Model=Qwen2.5-Math-7B2025.0740.8
Prefix-RFTBase Model=Qwen2.5-Math-7B2025.0740.3
SFT+RFTBase Model=Qwen2.5-Math-7B2025.0739.7
RFTBase Model=Qwen2.5-Math-7B2025.0739.3
LUFFYBase Model=Qwen2.5-Math-7B2025.0737.5
RL w/ SFT LossBase Model=Qwen2.5-Math-7B2025.0734.9
Oat-ZeroBase Model=Qwen2.5-Math-7B2025.0734.6
UFTBase Model=Qwen2.5-Math-7B2025.0733.8
ReLIFTBase Model=Qwen2.5-Math-7B2025.0733.8
SimpleRL-ZeroBase Model=Qwen2.5-Math-7B2025.0725
Qwen2.5-Math-7B2025.077.4