InternLM 开源 EndoCoT：在扩散模型中扩展内生 CoT 推理

InternLM; EndoCoT

Repository navigation

NotificationsYou must be signed in to change notification settings; Fork0 (0); Star49 (49)You must be signed in to star a repository

[ECCV 2026] An official implementation of "EndoCoT". Scaling endogenous Chain-of-Thought (CoT) reasoning in diffusion models for complex structured generation.

Repository files navigation

README; Contributing; MIT license

📝TODO

Open source the training code; Open source the training data; Open source the main task ckpt; Open source the edit model ckpt; Refactor the codebase for better usability and maintainability

🎉 [2026/6/26] Our paper has been accepted to ECCV 2026!; 🚀 [2026/3/18] We have released the EndoCoT-Data; 🚀 [2026/3/12] We have released the EndoCoT repository and ckpts

🌟Highlight

EndoCoT is a reasoning paradigm for diffusion models that enables step-by-step inference. It outperforms conventional training methods on Qwen-Image-Edit-2511

And provide transparent, intermediate reasoning trajectories

Following the configuration of Diffthinker, we provide a customized checkpoint for Qwen-Image-Edit. This checkpoint has been merged from the original safetensors to ensure compatibility with*Diffsynth-Studio* training. Please use the checkpoint provided in this repository instead of the official version for correct loading and inference.

Download the ckpt:

You may find our pretrained weights at: EndoCoT; Test Single Case; Eval Our Ckpt

We follow the exact same setting as Diffthinker

Training

Since the metadata uses relative paths, please ensure the dataset files are placed in the same directory as metadata.csv

Download the datasets & metadata.csv

You may find our training data at: EndoCoT dataset; Train your model

How to change the latent reasoning steps?

Note on Customization: Since the current implementation is straightforward, you can only manually adjust the latent reasoning steps in DiffSynth-Studio/diffsynth/pipelines/qwen_image.py:

Line 442: Modify infer_steps. Line 471: Modify training_steps.

We plan to optimize this in future releases.

[ECCV 2026] An official implementation of "EndoCoT". Scaling endogenous Chain-of-Thought (CoT) reasoning in diffusion models for complex structured generation.