List your company
ResearchFormerly WizWand

Diagram Understanding on AI2D lite

82.8AccuracyPVM-8B (SFT + GRPO)
76.5678.1879.881.42May 1, 2026
Updated 6mo ago

Evaluation results

MethodLinks
PVM-8B (SFT + GRPO)Backbone=8B, Training Strategy=SFT + GRPO, Model Architecture=Persistent Visual Memory2026.0582.8
Euclid-8BBackbone=8B, Training Strategy=RL-tuned2026.0582.6
PEARL-8BBackbone=8B, Training Strategy=RL-tuned2026.0581.8
OneThinker-8BBackbone=8B, Training Strategy=RL-tuned2026.0581.4
Qwen3-VL-8B (LoRA-SFT + GRPO)Backbone=8B, Training Strategy=LoRA-SFT + GRPO2026.0581
PVM-4B (SFT + GRPO)Backbone=4B, Training Strategy=SFT + GRPO, Model Architecture=Persistent Visual Memory2026.0581
PVM-8B (SFT)Backbone=8B, Training Strategy=SFT, Model Architecture=Persistent Visual Memory2026.0580.8
PVM-4B (SFT)Backbone=4B, Training Strategy=SFT, Model Architecture=Persistent Visual Memory2026.0580
Qwen3-VL-8B-InstructBackbone=8B2026.0579.8
Qwen3-VL-8B (LoRA-SFT)Backbone=8B, Training Strategy=LoRA-SFT2026.0579.8
CoMemoTraining Strategy=Visual Injection2026.0579.6
Qwen3-VL-8B (SFT + GRPO)Backbone=8B, Training Strategy=SFT + GRPO2026.0579.6
Qwen3-VL-4B (LoRA-SFT)Backbone=4B, Training Strategy=LoRA-SFT2026.0579.2
Qwen3-VL-8B (SFT)Backbone=8B, Training Strategy=SFT2026.0579
MemVRTraining Strategy=Visual Injection2026.0578.8
ICoTTraining Strategy=Visual Injection2026.0578.6
Qwen3-VL-4B (SFT + GRPO)Backbone=4B, Training Strategy=SFT + GRPO2026.0578.6
Qwen3-VL-4B-InstructBackbone=4B2026.0578.4
Qwen3-VL-4B (SFT)Backbone=4B, Training Strategy=SFT2026.0577.6
Qwen3-VL-4B (LoRA-SFT + GRPO)Backbone=4B, Training Strategy=LoRA-SFT + GRPO2026.0576.8