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

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

The official implementation is hosted on GitHub:

👉 https://github.com/NiuHechang/Calibrated_Adaptive_Length

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

Hengchang Liu, Zhao Yang, Bing Su

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 (Calibrated Adaptive Length) 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.

The method is evaluated on multiple infilling benchmarks including:

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

If you find this work useful, please cite: