Logics-Parsing - Alibaba's open-source end-to-end document parsing model

What is Logics-Parsing?

Logics-Parsing is an open-source end-to-end document parsing model from Alibaba, based on Qwen2.5-VL-7B. It optimizes document layout analysis and reading order inference through reinforcement learning, converting PDF images into structured HTML output. It supports various content types, including plain text, mathematical formulas, tables, chemical formulas, and handwritten Chinese characters. It employs a two-stage training process: the first stage is supervised fine-tuning, learning to generate structured output; the second stage is layout-centric reinforcement learning, optimizing text accuracy, layout localization, and reading order. It performs exceptionally well on the LogicsParsingBench benchmark, particularly outperforming other methods in parsing plain text, chemical structures, and handwritten content.

Main functions of Logics-Parsing

• End-to-end document parsingLogics-Parsing can directly convert PDF images into structured HTML output, supporting various content types, including plain text, mathematical formulas, tables, chemical formulas, and handwritten Chinese characters.
• Reinforcement learning optimizationA two-stage training method is adopted. In the first stage, structured output is generated through supervised fine-tuning learning. In the second stage, layout-centered reinforcement learning optimizes text accuracy, layout positioning, and reading order.
• High performanceIn the LogicsParsingBench benchmark, Logics-Parsing performs exceptionally well across multiple document types, particularly outperforming other methods in parsing plain text, chemical structures, and handwritten content.
• Wide range of applicationsSuitable for various complex document types such as academic papers, complex multi-column documents, newspapers, and posters, it can handle multi-column layouts, mathematical formulas, and chemical structures.

Technical principles of Logics-Parsing

• Based on Qwen2.5-VL-7BLogics-Parsing is based on the powerful Qwen2.5-VL-7B model, inheriting its advantages in vision and language processing.
• Two-stage trainingThe first stage is supervised fine-tuning, where the model learns to generate structured HTML output; the second stage is layout-centric reinforcement learning, which optimizes the model through three reward components: text accuracy, layout localization, and reading order.
• Reinforcement learning optimizationThrough reinforcement learning, the model can better understand the layout of documents and the logical order of content, generating more accurate structured output.
• Structured HTML outputThe model can convert document images into a structured HTML format, preserving the document's logical structure. Each content block has a category, bounding box coordinates, and OCR text labels.
• Advanced Content RecognitionThe model can accurately recognize complex scientific formulas, chemical structures, and handwritten Chinese characters, and convert chemical structures into the standard SMILES format.
• Automatically remove irrelevant elementsThe model can automatically identify and filter out irrelevant elements such as headers and footers, allowing you to focus on the core content of the document.

Project address for Logics-Parsing

• Github repositoryhttps://github.com/alibaba/Logics-Parsing
• HuggingFace model libraryhttps://huggingface.co/Logics-MLLM/Logics-Parsing
• arXiv technical paperhttps://arxiv.org/pdf/2509.19760

Application scenarios of Logics-Parsing

• Analysis of academic papersIt can process academic papers containing multi-column layouts, mathematical formulas, and chemical structures, accurately extracting key information and generating structured output.
• Complex multi-column documentsSuitable for complex multi-column documents such as newspapers and posters, effectively parsing the layout and extracting content.
• Handwritten document recognitionIt supports the recognition and parsing of handwritten Chinese characters and can be applied to scenarios such as handwritten notes and exam papers.
• Chemical document processingIt accurately identifies chemical formulas and converts them into the standard SMILES format, making it suitable for document parsing in the field of chemistry.
• Mathematical document parsingIt can process documents containing complex mathematical formulas, such as math textbooks and research papers, and provide accurate formula analysis.
• Multilingual document supportIt supports document parsing in multiple languages and is suitable for internationalized document processing scenarios.