[논문리뷰] Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and GenerationYichen Zhang이 arXiv에 게시한 'Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation' 논문에 대한 자세한 리뷰입니다.#Review#Unified multimodal model#Visual generation and comprehension#Unified vision encoder#Cascaded flow matching#Token compression2026년 3월 15일댓글 수 로딩 중