List your company

Vision-Dependent Multimodal Math Reasoning on MathVerseV Vision-centric

64.97Avg@8 AccuracyPAPO_G
44.627649.908855.1960.4712Jul 8, 2025

Evaluation results

MethodLinks
PAPO_GBackbone=Qwen2.5-VL, Model Size=7B2025.0764.97
PAPO_DBackbone=Qwen2.5-VL, Model Size=7B2025.0764.87
GRPOBackbone=Qwen2.5-VL, Model Size=7B2025.0761.71
PAPO_DBackbone=Qwen2.5-VL, Model Size=3B2025.0757.72
PAPO_GBackbone=Qwen2.5-VL, Model Size=3B2025.0753.95
DAPOBackbone=Qwen2.5-VL, Model Size=3B2025.0753.09
GRPOBackbone=Qwen2.5-VL, Model Size=3B2025.0752.26
DAPOBackbone=Qwen2.5-VL, Model Size=7B2025.0751.04
PAPO_GBackbone=Qwen3-VL (thinking), Model Size=2B2025.0750.05
GRPOBackbone=Qwen3-VL (thinking), Model Size=2B2025.0745.41