
UniCode proposes a novel approach within multimodal large language models (MLLMs) that learns a unified codebook to efficiently tokenize visual, text, and potentially other types of signals.


coremsg

unicode-learning-a-unified-codebook-for-multimodal-large-language-models


UniCode: Learning a Unified Codebook for Multimodal Large Language Models


title_rewrite


UniCode proposes a unified codebook for multimodal large language models to enhance visual and text tokenization efficiency.
