UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm

Benchmark Model Rank Results
multimodal-reasoning-on-visulogicUVU 3B–Accuracy: 26.4
science-question-answering-on-scienceqaUVU 3B–Avg. Accuracy: 91.3
vision-language-understanding-on-mmstarUVU 3B–Accuracy: 55.0
visual-question-answering-vqa-on-blinkUVU 3B–Accuracy: 52.8