Official Implementation of DINGO: Constrained Inference for Diffusion LLMs published at the NeurIPS 2025.
This is the visualization of the DINGO workflow, which seamlessly handles any regular expression, including those used in JSON Schema definitions. DINGO is optimal and always draws the highest probability string that is structurally valid.
@misc{suresh2025dingoconstrainedinferencediffusion,
title={DINGO: Constrained Inference for Diffusion LLMs},
author={Tarun Suresh and Debangshu Banerjee and Shubham Ugare and Sasa Misailovic and Gagandeep Singh},
year={2025},
eprint={2505.23061},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2505.23061},
}