0.9B小模型实现高效文档理解，多标记预测提速解码
GLM-OCR Technical Report
- 采用视觉-语言双分支架构，0.4B+0.5B参数分工协作
- 多标记预测机制使每步输出多个词元，解码速度显著提升
- 支持边缘部署与大规模生产，适合实际场景落地
GLM-OCR 是一个针对真实世界文档理解设计的高效小型多模态模型，总参数量为 0.9B，由 0.4B 参数的 CogViT 视觉编码器和 0.5B 参数的 GLM 语言解码器构成，兼顾计算效率与识别性能。为解决标准自回归解码在确定性 OCR 任务中的低效问题，该模型引入多标记预测（MTP）机制，每步预测多个词元，显著提升解码吞吐量，同时通过共享参数保持低内存开销。系统层面采用两阶段流水线：先由 PP-DocLayout-V3 完成版面分析，再并行执行区域级识别。在公开基准与工业场景的广泛评估中，GLM-OCR 在文档解析、文本与公式转录、表格结构恢复及关键信息提取任务上达到竞争力或领先水平。其紧凑架构与结构化生成能力，使其适用于资源受限的边缘部署及大规模生产系统。
原文摘要 · Abstract (English)
GLM-OCR is an efficient 0.9B-parameter compact multimodal model designed for real-world document understanding. It combines a 0.4B-parameter CogViT visual encoder with a 0.5B-parameter GLM language decoder, achieving a strong balance between computational efficiency and recognition performance. To address the inefficiency of standard autoregressive decoding in deterministic OCR tasks, GLM-OCR introduces a Multi-Token Prediction (MTP) mechanism that predicts multiple tokens per step, significantly improving decoding throughput while keeping memory overhead low through shared parameters. At the system level, a two-stage pipeline is adopted: PP-DocLayout-V3 first performs layout analysis, followed by parallel region-level recognition. Extensive evaluations on public benchmarks and industrial scenarios show that GLM-OCR achieves competitive or state-of-the-art performance in document parsing, text and formula transcription, table structure recovery, and key information extraction. Its compact architecture and structured generation make it suitable for both resource-constrained edge deployment and large-scale production systems.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。