Logics-MLLM/Logics-Parsing-v2

Logics-Parsing-v2: Advanced Document Understanding

Logics-Parsing-v2 is an evolution of the Logics-Parsing model, developed by Logics-MLLM, specifically designed for comprehensive document parsing. This 4 billion parameter model extends its capabilities to handle highly complex documents and introduces Parsing-2.0 scenarios, enabling structured interpretation of diverse content types.

Key Capabilities

• End-to-End Document Processing: Performs recognition and parsing for various document elements within a single model, handling complex layouts and text-dense documents like newspapers and magazines with high precision.
• Advanced Content Recognition: Delivers accurate and structured parsing of tables and scientific formulas. Its Parsing-2.0 feature natively supports parsing of flowcharts, music sheets (using ABC notation), and pseudocode blocks (using Mermaid for flowcharts).
• Rich, Structured HTML Output: Converts documents into concise HTML, preserving content, spatial layouts, and semantic hierarchy, with specialized formats for structured elements.

Performance and Benchmarks

Logics-Parsing-v2 demonstrates state-of-the-art performance across multiple benchmarks:

• Achieves an overall score of 82.16 on the in-house LogicsDocBench, a comprehensive evaluation benchmark comprising 900 PDF pages covering traditional Parsing-1.0 and new Parsing-2.0 scenarios (STEM documents, complex layouts, and Parsing-2.0 content).
• Sets top records on the public OmniDocBench-v1.5 with an overall score of 93.23, outperforming other approaches.

Should you use this for your use case?

This model is ideal for applications requiring highly accurate and structured extraction of information from diverse and complex documents. If your use case involves parsing scientific papers, technical manuals, financial reports, or even specialized content like musical scores and flowcharts, Logics-Parsing-v2 offers robust capabilities to convert these into actionable, structured data.