internlm/EndoCoT-Data

---
language:
- en
license: cc-by-nc-4.0
task_categories:
- image-to-image
datasets:
- internlm/EndoCoT-Data
base_model:
- Qwen/Qwen-Image-Edit-2511
---

<p align="center"> <img src="fig/banner.svg" alt="EndoCoT" width="900"/> </p>

<p align="center">
<a href="https://github.com/InternLM/EndoCoT"><img src="https://img.shields.io/github/stars/InternLM/EndoCoT?style=flat-square&logo=github&label=Stars&color=FFB300"></a>
<a href="https://github.com/InternLM/EndoCoT/forks"><img src="https://img.shields.io/github/forks/InternLM/EndoCoT?style=flat-square&logo=github&label=Forks&color=2196F3"></a>
<a href="https://github.com/InternLM/EndoCoT/issues"><img src="https://img.shields.io/github/issues/InternLM/EndoCoT?style=flat-square&logo=github&label=Issues&color=4CAF50"></a>
<a href="https://github.com/InternLM/EndoCoT/blob/main/LICENSE"><img src="https://img.shields.io/github/license/InternLM/EndoCoT?style=flat-square&label=License&color=9C27B0"></a>
<br>
<a href="https://arxiv.org/abs/2603.12252"><img src="https://img.shields.io/badge/Paper-arXiv-B31B1B?style=flat-square"></a>
<a href="https://internlm.github.io/EndoCoT/"><img src="https://img.shields.io/badge/Homepage-Project-blue?style=flat-square"></a>
<a href="https://huggingface.co/internlm/EndoCoT"><img src="https://img.shields.io/badge/Model-HuggingFace-yellow?style=flat-square"></a>
<a href="https://huggingface.co/datasets/internlm/EndoCoT-Data"><img src="https://img.shields.io/badge/Dataset-HuggingFace-orange?style=flat-square"></a>
<br>
<br>
<img src="fig/teaser.jpg" alt="Teaser" width="100%" style="border-radius: 10px; box-shadow: 0 6px 20px rgba(0,0,0,0.2);">
</p>

# EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models

This repository contains the training data for **EndoCoT**, a novel framework that activates the reasoning potential of Multimodal Large Language Models (MLLMs) within diffusion frameworks through an iterative thought guidance module.

- **Paper:** [EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models](https://arxiv.org/abs/2603.12252)
- **Project Page:** [https://internlm.github.io/EndoCoT/](https://internlm.github.io/EndoCoT/)
- **Repository:** [https://github.com/InternLM/EndoCoT](https://github.com/InternLM/EndoCoT)

## 🌟 Highlights

- **EndoCoT** is a reasoning paradigm for diffusion models that enables step-by-step inference.
- It outperforms conventional training methods on complex tasks like Maze, TSP, VSP, and Sudoku.
- Provides transparent, intermediate reasoning trajectories.

## ⚡ Quick Start

### Setup environment

```bash
git clone https://github.com/InternLM/EndoCoT
cd EndoCoT
conda create -n EndoCoT python=3.10
conda activate EndoCot
pip install -r requirements.txt
```

### Sample Usage (Inference)

To test a single case using the codebase:

```bash
cd test
python test.py \
--task Maze \
--model_root /path/to/merged_ckpts \
--lora_path /path/to/your_lora_weight.safetensors \
--input_image ./data/sudoku_sample.png \
--output_dir ./outputs/sudoku_results
```

### Training

1. Download the datasets & `metadata.csv` and ensure they are placed in the same directory.
2. Run the training scripts:

```bash
cd DiffSynth-Studio
bash add/Maze/stage1.sh
python change_ckpt_prefix.py --src /path/to/the/Maze/save/dir/Maze_stage1
bash add/Maze/stage2.sh
python change_ckpt_prefix.py --src /path/to/the/Maze/save/dir/Maze_stage2
```

## 📰 News

- 🚀 [2026/3/12] We have released the EndoCoT [repository](https://github.com/InternLM/EndoCoT) and [ckpts](https://huggingface.co/internlm/EndoCoT).

## 📖 Citation

```
@article{dai2026endocot,
title={EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models},
author={Dai, Xuanlang and Zhou, Yujie and Xing, Long and Bu, Jiazi and Wei, Xilin and Liu, Yuhong and Zhang, Beichen and Chen, Kai and Zang, Yuhang},
journal={arXiv preprint arXiv:2603.12252},
year={2026}
}
```

## ⚖️ License

The code in the associated repository is licensed under the **MIT License**. The dataset is licensed under the **CC BY-NC 4.0 License**.