UniRec-0.1B : Unified Text and Formula Recognition with 0.1B Parameters
Yongkun Du 1, Zhineng Chen 1, † \dagger, Yazhen Xie 1, Weikang Bai 1, Hao Feng 2, Wei Shi 2, Yuchen Su 1, Can Huang 2, Yu-Gang Jiang 1, † \dagger · 2025
Read UniRec-0.1B : Unified Text and Formula Recognition with 0.1B Parameters by Yongkun Du 1, Zhineng Chen 1, † \dagger, Yazhen Xie 1, Weikang Bai 1, Hao Feng 2, Wei Shi 2, Yuchen Su 1, Can Huang 2, Yu-Gang Jiang 1, † \dagger (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-2512-21095.txt.
Description
Yongkun Du, Yu-Gang Jiang, and seven co-authors shrink document recognition to a tenth of a billion parameters without giving up breadth. Their model UniRec reads both text and mathematical formulas, at every scale from single characters to whole documents. That job lately belongs to large vision-language models too heavy for many deployments. Training rests on a new forty-million-sample dataset mixing text, formulas, and both together.
Two obstacles get named and engineered around. Structure varies wildly across recognition levels; hierarchical supervision answers that by teaching the structure explicitly. Text and formula content also entangle semantically. A tokenizer answers that by keeping their representations separate. On their bilingual benchmark and public ones, the small model beats general vision-language models and specialist parsers while running two to nine times faster.
Catalog record
| CREATOR | Yongkun Du 1, Zhineng Chen 1, † \dagger, Yazhen Xie 1, Weikang Bai 1, Hao Feng 2, Wei Shi 2, Yuchen Su 1, Can Huang 2, Yu-Gang Jiang 1, † \dagger |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-21095