List your company

Mathematical Reasoning on MATH 500 (avg@1)

81.6Avg@1 ScoreVIMPO
52.89660.34867.875.252Jun 18, 2026

Evaluation results

MethodLinks
VIMPOReward condition=Clean, Training steps=Full training2026.0681.6
Naive GRPOReward condition=Clean, Training steps=Full training2026.0680.4
GRPOReward condition=Clean, Training steps=Full training2026.0679.6
VIMPOReward condition=25% reward flipping, Training steps=200 steps2026.0678.2
GRPOReward condition=25% reward flipping, Training steps=200 steps2026.0675.2
Qwen3-4B-BaseReward condition=Clean, Training steps=Full training2026.0654