Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation

Benchmark Model Rank Results
visual-question-answering-vqa-on-seed-seedbenchCheers (1.5B)#21Accuracy: 71.7
text-to-image-generation-on-dpg-benchCheers (1.5B)#65DPG-Bench Score: 83.48
chart-question-answering-on-chartqaCheers (1.5B)#811:1 Accuracy: 75.7
visual-question-answering-vqa-on-realworldqaCheers (1.5B)#82Accuracy: 60.9
vision-language-understanding-on-mmstarCheers (1.5B)#92Accuracy: 50.9
multimodal-reasoning-on-mathvistaCheers (1.5B)#106Accuracy: 50.5
text-to-image-generation-on-genevalCheers#112Overall: 0.78
multimodal-reasoning-on-mmmuCheers (1.5B)#129Accuracy: 36.0