List your company
ResearchFormerly WizWand

Vision-Language Modeling on WeMath

82.6AccuracyZPPO
53.27260.88668.576.114Jun 16, 2026
Updated 4mo ago

Evaluation results

MethodLinks
ZPPOModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.0682.6
GRPOModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.0680.7
On-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.0680.6
GRPOModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.0680.4
Off-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.0679.5
Off-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.0679.3
On-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.0679.3
Qwen3.5Model Scale=2B, Prompt Replay Buffer Augmentation=false2026.0677.9
ZPPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.0671.7
GRPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.0662.5
GRPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.0660.6
On-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.0658.6
On-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.0657.7
Off-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.0656.4
Off-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.0655.9
Qwen3.5Model Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.0654.4