DreamOn: Diffusion Language Models For Code Infilling Beyond Fixed-size Canvas
Zirui Wu, Lin Zheng, Zhihui Xie, Jiacheng Ye, Jiahui Gao, Shansan Gong, Yansong Feng, Zhenguo Li, Wei Bi, Guorui Zhou, Lingpeng Kong · 2026
Read DreamOn: Diffusion Language Models For Code Infilling Beyond Fixed-size Canvas by Zirui Wu, Lin Zheng, Zhihui Xie, Jiacheng Ye, Jiahui Gao, Shansan Gong, Yansong Feng, Zhenguo Li, Wei Bi, Guorui Zhou, Lingpeng Kong (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-01326.txt.
Description
Zirui Wu and co-authors build DreamOn, a diffusion-word plan for code fill-ins of any length. Diffusion word models fill gaps in any order with no special prompts, but they demand a fixed mask length, so a wrong mask wrecks fill quality. DreamOn adds two length-control states letting the model stretch or shrink output from its own guesses, with tiny train-objective tweaks and no net changes. Built on two 7B diffusion coders, it matches top autoregressive fillers and oracle-length scores on fill tests.
Catalog record
| CREATOR | Zirui Wu, Lin Zheng, Zhihui Xie, Jiacheng Ye, Jiahui Gao, Shansan Gong, Yansong Feng, Zhenguo Li, Wei Bi, Guorui Zhou, Lingpeng Kong |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38131 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01326