Spaces:
Sleeping
Sleeping
File size: 3,227 Bytes
0533780 363e6f1 0533780 363e6f1 0533780 886ad57 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 0533780 363e6f1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 | ---
title: GLM OCR
emoji: 📄
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 7860
pinned: false
---
# GLM-OCR — Self-Hosted OCR Engine
> A full-stack portfolio project: self-hosted OCR backend powered by **zai-org/GLM-OCR**, a 0.9B-param vision-language model ranked #1 on OmniDocBench V1.5.
### 🔗 [Live Demo → https://huggingface.co/spaces/Sam20202/GLMOCR_Text_extraction](https://huggingface.co/spaces/Sam20202/GLMOCR_Text_extraction)
---
## What is GLM-OCR?
GLM-OCR is a state-of-the-art open-source OCR model from
**zai-org** (arXiv:2603.10910).
Unlike traditional OCR (Tesseract, etc.) it uses a **vision encoder + language model** architecture:
```
Image → [CogViT Vision Encoder] → [GLM-0.5B LM Backbone] → Text
```
It handles:
- Plain text from documents, screenshots, photos
- **Tables** (preserved structure)
- **Mathematical equations** (LaTeX output)
- **Code blocks** (syntax preserved)
- **Multilingual** text
- **Handwriting**
---
## Project Structure
```
GLMOCR_Text_extraction/
├── main.py # FastAPI server — routes, CORS, request handling
├── ocr_engine.py # Model loading, inference, OcrResult dataclass
├── requirements.txt # Python dependencies
├── frontend/
│ └── index.html # Single-file frontend (served by FastAPI)
├── Extension/ # Browser extension (Chrome/Edge)
├── Dockerfile # HF Spaces deployment
├── docker-compose.yml # Local Docker deployment
└── README.md
```
## Architecture Diagram
```
Browser
│
│ POST /ocr (multipart image + mode)
▼
┌─────────────────────────────────────┐
│ FastAPI (main.py) │
│ ─ CORS middleware │
│ ─ file validation (type, size) │
│ ─ session metrics │
└──────────────┬──────────────────────┘
│ image_bytes, mode
▼
┌─────────────────────────────────────┐
│ GlmOcrEngine (ocr_engine.py) │
│ ─ PIL validation & RGB conversion │
│ ─ writes temp PNG to disk │
│ ─ processor.apply_chat_template() │
│ ─ returns OcrResult dataclass │
└──────────────┬──────────────────────┘
│ (torch.inference_mode)
▼
┌─────────────────────────────────────┐
│ GLM-OCR 0.9B Model │
│ zai-org/GLM-OCR │
│ ─ Vision encoder (CogViT) │
│ ─ LM backbone (GLM-0.5B) │
│ ─ Runs on CUDA / CPU / MPS │
└─────────────────────────────────────┘
```
## References
- Paper: [GLM-OCR](https://arxiv.org/abs/2603.10910)
- Model: [zai-org/GLM-OCR](https://huggingface.co/zai-org/GLM-OCR)
|