UniRec-0.1B: Unified Text and Formula Recognition with 0.1B Parameters

2025-12-24

AI Summary

Existing vision-language models (VLMs) suffer from excessive parameter counts and high computational overhead, hindering efficient multi-granularity joint recognition of text and mathematical formulas in documents. To address this, we propose UniRec—a lightweight unified recognition model with only 0.1B parameters. Our method introduces a hierarchical supervision training paradigm and a semantic-decoupled tokenizer to explicitly disentangle textual and formulaic semantics while capturing structural variability. We construct UniRec40M, a large-scale, 40-million-sample hybrid dataset spanning diverse document domains. UniRec employs a lightweight vision-language modeling framework coupled with a multi-granularity sequence recognition architecture. On bilingual (Chinese/English), multi-domain document understanding benchmarks, it consistently outperforms state-of-the-art VLMs and domain-specific parsing models, achieving 2–9× faster inference speed. This advancement significantly enhances the practicality of on-device document understanding.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Text and formulas constitute the core informational components of many documents. Accurately and efficiently recognizing both is crucial for developing robust and generalizable document parsing systems. Recently, vision-language models (VLMs) have achieved impressive unified recognition of text and formulas. However, they are large-sized and computationally demanding, restricting their usage in many applications. In this paper, we propose UniRec-0.1B, a unified recognition model with only 0.1B parameters. It is capable of performing text and formula recognition at multiple levels, including characters, words, lines, paragraphs, and documents. To implement this task, we first establish UniRec40M, a large-scale dataset comprises 40 million text, formula and their mix samples, enabling the training of a powerful yet lightweight model. Secondly, we identify two challenges when building such a lightweight but unified expert model. They are: structural variability across hierarchies and semantic entanglement between textual and formulaic content. To tackle these, we introduce a hierarchical supervision training that explicitly guides structural comprehension, and a semantic-decoupled tokenizer that separates text and formula representations. Finally, we develop a comprehensive evaluation benchmark covering Chinese and English documents from multiple domains and with multiple levels. Experimental results on this and public benchmarks demonstrate that UniRec-0.1B outperforms both general-purpose VLMs and leading document parsing expert models, while achieving a 2-9$ imes$ speedup, validating its effectiveness and efficiency. Codebase and Dataset: https://github.com/Topdu/OpenOCR.

Problem

• Unified recognition of text and formulas in documents
• Reducing model size and computational demands for efficiency
• Addressing structural and semantic challenges in mixed content

Innovation

• Lightweight 0.1B-parameter model for unified text and formula recognition
• Hierarchical supervision training for multi-level structural comprehension
• Semantic-decoupled tokenizer separating text and formula representations