List your company

Multi-subject Knowledge on MMLU-Redux

67AccuracyReLIFT
46.61651.90857.262.492Jun 9, 2025

Evaluation results

MethodLinks
ReLIFT2025.06671,511
RL2025.0665.7863
LUFFY2025.0665.21,516
SFT then RL(v2)2025.06621,501
SFT2025.0660.12,479
RL w/ SFT Loss2025.0659.12,514
SFT then RL(v1)2025.0656.21,217
Qwen-Math-Instruct2025.0648.13,628
Qwen-Math2025.0647.4555