---
license: apache-2.0
---

# Diffusion LMs Can Approximate Optimal Infilling Lengths Implicitly

[![arXiv](https://img.shields.io/badge/Paper-arXiv-b31b1b.svg)](https://arxiv.org/abs/2602.00476)
[![GitHub](https://img.shields.io/badge/Code-GitHub-black.svg)](https://github.com/NiuHechang/Calibrated_Adaptive_Length)

This repository is the Hugging Face project page for the paper:

**Diffusion LMs Can Approximate Optimal Infilling Lengths Implicitly**

We propose **CAL (Calibrated Adaptive Length)**, a training-free framework that enables Diffusion Language Models to approximate optimal infilling lengths without additional training.

---

## 🔗 Code Repository

The official implementation is hosted on GitHub:

👉 **[https://github.com/NiuHechang/Calibrated_Adaptive_Length](https://github.com/NiuHechang/Calibrated_Adaptive_Length)**

This Hugging Face repository only serves as a landing page linking to the official codebase and paper.

---

## 📄 Paper

**Diffusion LMs Can Approximate Optimal Infilling Lengths Implicitly**

Hengchang Liu, Zhao Yang, Bing Su

[https://arxiv.org/abs/2602.00476](https://arxiv.org/abs/2602.00476)

---

## 📝 Abstract (from the paper)

Diffusion language models (DLMs) provide a bidirectional generation framework naturally suited for infilling, yet their performance is constrained by the pre-specified infilling length. In this paper, we reveal that DLMs possess an inherent ability to discover the correct infilling length. We identify two key statistical phenomena in the first-step denoising confidence: a local *Oracle Peak* that emerges near the ground-truth length and a systematic *Length Bias* that often obscures this signal. By leveraging this signal and calibrating the bias, our training-free method **CAL** (**C**alibrated **A**daptive **L**ength) enables DLMs to approximate the optimal length through an efficient search before formal decoding. Empirical evaluations demonstrate that CAL improves Pass@1 by up to 47.7\% over fixed-length baselines and 40.5% over chat-based adaptive methods in code infilling, while boosting BLEU-2 and ROUGE-L by up to 8.5% and 9.9% in text infilling. These results demonstrate that CAL paves the way for robust DLM infilling without requiring any specialized training.

## 🧠 Evaluation

The method is evaluated on multiple infilling benchmarks including:

* HumanEval-Infilling (Code)
* ROCStories (Text)
* CSAbstracts (Text)
* Yelp Reviews (Text)

For full implementation details and experiment setup, please refer to the GitHub repository.

---

## 📖 Citation

If you find this work useful, please cite:

```bibtex
@misc{liu2026diffusionlmsapproximateoptimal,
      title={Diffusion LMs Can Approximate Optimal Infilling Lengths Implicitly}, 
      author={Hengchang Liu and Zhao Yang and Bing Su},
      year={2026},
      eprint={2602.00476},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2602.00476}, 
}
```