Partition Generative Modeling: Masked Modeling Without Masks
Justin Deschenaux, Lan Tran, Caglar Gulcehre · 2025
Read Partition Generative Modeling: Masked Modeling Without Masks by Justin Deschenaux, Lan Tran, Caglar Gulcehre (2025) 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-2505-18883.txt.
Description
Two families of models write text in different ways: one fills in blanked-out words in any order, the other writes one word at a time from the left. Justin Deschenaux, Lan Tran, and Caglar Gulcehre blend the two in their 2025 paper. Their model splits the text into two groups that cannot see each other, and learns to predict each group from the other, so no blank tokens are ever needed.
On a large web text set, the new model generated samples five times faster than a leading masked model, with better quality scores. On images it matched a rival's quality while running far faster, and it still works with existing sampling tools.
Catalog record
| CREATOR | Justin Deschenaux, Lan Tran, Caglar Gulcehre |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-18883