GLM-OCR
A multimodal OCR model for reading complex documents and turning images of text into usable text. It targets accurate and fast document understanding.
Share on XLicense: Apache-2.0
Overview
GLM-OCR is a multimodal OCR model for understanding complex documents. It combines a visual encoder, a small cross-modal connector and a GLM-0.5B language decoder, and uses a two-stage pipeline of layout analysis followed by parallel recognition. It turns document images into text, including formulas and tables, and can be deployed with vLLM, SGLang or Ollama.
Key features
- Two-stage layout analysis and parallel recognition
- Handles formulas, tables and information extraction
- Only 0.9B parameters
- Deployable with vLLM, SGLang and Ollama
- Aimed at complex tables, code-heavy documents and seals
Best for
Teams that need to read complex real-world documents into text and want a compact model they can host themselves.
- Upstream
- zai-org/GLM-OCR
- Fork on GitHub
- Guo-astro/GLM-OCR
- Upstream stars
- 7.5k
- Category
- Documents, media and content
- Language
- Python
- License
- Apache-2.0
- Forked
- 2026-03-17
- Sync status
- In syncLast synced 2026-09-29
More in Documents, media and content
A Markdown-based typesetting system that turns plain text into papers, presentations, websites, books and knowledge bases. It adds scripting features to Markdown for richer documents.
Forked 2026-09-28Last synced 2026-09-29License: GPL-3.0Documents, media and contentGitHub
A fast library that converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF files into clean Markdown. It has Node.js and Python bindings, useful for feeding documents to text tools.
Forked 2026-09-28Last synced 2026-09-29License: MITDocuments, media and contentGitHub
A starter that creates a complete Astro documentation site in your repo as editable files. It is built so both human readers and AI agents can use the docs.
Forked 2026-09-28Last synced 2026-09-29License: MITDocuments, media and contentGitHub
A PowerPoint alternative where a whole slide deck is a single HTML file containing its own viewer, presenter and editor. You can open, edit and present it in any browser.
Forked 2026-09-26Last synced 2026-09-29License: MITDocuments, media and contentGitHub