Liang Wan

CoLT: Teaching Multi-Modal Models to Think with Chain of Latent Thoughts

Lianyu Hu, Shengqian Qin, Zeqin Liao, Qing Guo, Liang Wan, Wei Feng, et al.

Chain-of-thought (CoT) reasoning has enabled multi-modal large language models (MLLMs) to tackle complex visual reasoning tasks by generating explicit intermediate reasoning steps in natural language. However, this text-based reasoning paradigm is inherently slow at inference tim…Read more