CoI: Chain of Images for Intuitively Reasoning

This repository is the official implementation of "Chain of Images for Intuitively Reasoning". Welcome to discuss this paper with me on 知乎. Comparing to CoT which utlize language as an intermedia for solving complex task, Chain of Images (CoI) convert complex language reasoning problems to simple pattern recognition by generating a series of images as intermediate representations.

Demo

Welcom to try the demo of CoI running on huggingface space.

Todo list

• Release the CoIEval datasest
• Release the checkpoint of SyMLLM framework
• Build CoI Benchmark

Citation

If you find CoI useful, please cite it in your publications.

@article{meng2023chain,
  title={Chain of Images for Intuitively Reasoning},
  author={Meng, Fanxu and Yang, Haotong and Wang, Yiding and Zhang, Muhan},
  journal={arXiv preprint arXiv:2311.09241},
  year={2023}
}