Program-Verified Self-Evolution for Vision-Language Models

Open paper
Benchmark Model Rank Results
visual-question-answering-vqa-on-seed-seedbenchVQS-2B (Qwen3-VL-2B-Instruct)#19Accuracy: 72.61
multimodal-reasoning-on-mmmuVQS-2B (Qwen3-VL-2B-Instruct)#103Accuracy: 46.78

Rank counts only results with a code link.