List your company

Multimodal Math Reasoning on MathVerse (avg@8 accuracy)

68.58Avg@8 AccuracyPAPO_D
47.15652.71858.2863.842Jul 8, 2025

Evaluation results

MethodLinks
PAPO_DBackbone=Qwen2.5-VL, Model Size=7B2025.0768.58
PAPO_GBackbone=Qwen2.5-VL, Model Size=7B2025.0768.43
GRPOBackbone=Qwen2.5-VL, Model Size=7B2025.0766.51
PAPO_DBackbone=Qwen2.5-VL, Model Size=3B2025.0760.51
PAPO_GBackbone=Qwen2.5-VL, Model Size=3B2025.0757.14
DAPOBackbone=Qwen2.5-VL, Model Size=3B2025.0756.25
DAPOBackbone=Qwen2.5-VL, Model Size=7B2025.0755.64
GRPOBackbone=Qwen2.5-VL, Model Size=3B2025.0755.25
PAPO_GBackbone=Qwen3-VL (thinking), Model Size=2B2025.0751.89
GRPOBackbone=Qwen3-VL (thinking), Model Size=2B2025.0747.98