The authors first trained a vector-quantized visual tokenizer, called VQ-VT, to encode intermediate visual states produced by tools such as cropping, zooming or image transformation.
Multimodal models learn when to look and how long to reason
L2T places visual and textual computation in one token sequence, then adapts both the reasoning mode and computation budget to each task.
Chinese Tech
Quanlin Chen · Xiuwei Chen · Kun Xiang · Zehua Ma · Chunjin Ye · Boqiang Guo · +2 more
Sun Yat-sen University · Shenzhen Loop Area Institute · Beijing Zhongguancun Academy · Huawei Technologies Co., Ltd · Harbin Institute of Technology
Research Digest··2 min read
Chen et al.
Why this paper
From Huawei Technologies Co., Ltd and 4 others · Part of Visual Token Efficiency, now 4 papers
In one line
L2T jointly optimizes the form and depth of computation for multimodal reasoning.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§