GLM-OCR-Demo

A Gradio-based demonstration for the zai-org/GLM-OCR multimodal OCR model. Supports text, formula, and table recognition from uploaded images, with outputs in plain text and markdown formats. Features custom HotPink theme, GPU acceleration, image orientation handling (EXIF transpose), and temporary file management for processing.

Features

• Recognition Types: Text Recognition, Formula Recognition, Table Recognition with predefined prompts.
• Image Handling: Supports upload/clipboard sources; auto-converts RGBA/LA/P modes to RGB; handles EXIF orientation.
• Outputs: Dual tabs for plain text and markdown rendering.
• Custom Theme: HotPinkTheme with responsive, animated styling via CSS.
• GPU Inference: Uses spaces.GPU decorator for efficient processing.
• Examples: 5 curated images for quick testing.
• Queueing: Up to 50 concurrent jobs.

Prerequisites

• Python 3.10 or higher.
• CUDA-compatible GPU (recommended for bfloat16 inference).
• Stable internet for initial model downloads from Hugging Face.

Installation

1. Clone the repository:

   git clone https://github.com/PRITHIVSAKTHIUR/GLM-OCR-Demo.git
   cd GLM-OCR-Demo
   

2. Install dependencies: First, install pre-requirements:

   pip install -r pre-requirements.txt
   

   Then, install main requirements:

   pip install -r requirements.txt
   

   pre-requirements.txt content:

   pip>=23.0.0
   

   requirements.txt content:

   flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.3/flash_attn-2.7.3+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl #skippable
   git+https://github.com/huggingface/transformers.git
   git+https://github.com/huggingface/accelerate.git
   git+https://github.com/huggingface/peft.git
   huggingface_hub
   sentencepiece
   opencv-python
   torch==2.6.0
   torchvision
   matplotlib
   markdown
   requests
   hf_xet
   spaces
   pillow
   gradio #@gradio6
   av
   

3. Start the application:

   python app.py
   

   The demo launches at http://localhost:7860.

Usage

1. Upload Image: Add an image via upload or clipboard.
2. Select Task: Choose Text, Formula, or Table recognition.
3. Recognize: Click "Recognize" to process.
4. View Outputs: Check results in Text or Markdown tabs.

Supported Tasks

Task	Prompt
Text	"Text Recognition:"
Formula	"Formula Recognition:"
Table	"Table Recognition:"

Examples

Input Image	Task
examples/1.jpg	Text
examples/4.jpg	Text
examples/5.webp	Text
examples/2.jpg	Formula
examples/3.jpg	Table

Troubleshooting

• Model Loading: First run downloads GLM-OCR; monitor console.
• Image Errors: Ensure valid RGB images; check console for processing issues.
• OOM: Use smaller images or reduce max_new_tokens (default 8192).
• No Output: Upload image first; select task.
• Flash Attention: Requires compatible CUDA; fallback if fails.

Contributing

Contributions welcome! Add new tasks to TASK_PROMPTS, enhance CSS, or improve processing. Submit pull requests via the repository.

Repository: https://github.com/PRITHIVSAKTHIUR/GLM-OCR-Demo.git

License

Apache License 2.0. See LICENSE for details. Built by Prithiv Sakthi. Report issues via the repository.