DeepSeek-OCR-2
DeepSeek's 3B document-understanding OCR — reads documents like a human
DeepSeek's 3B document-understanding OCR that reads documents like a human.
Translate documents, parse PDFs, and extract text with OCR.
Key features
- DeepSeek-OCR 2 exploring human-like visual encoding (Visual Causal Flow)
- Dynamic resolution: (0-6)x768x768 plus 1x1024x1024 crops
- Prompts to convert documents to markdown
- Runs via vLLM or Hugging Face Transformers
Good for
- Developers converting documents to markdown with an open VLM
- Researchers benchmarking on OmniDocBench v1.5
Stack & tags
PythonPyTorchvLLMTransformersocrdocument-aivlmmarkdownopen-weightstransformers
Quick start
git clone https://github.com/deepseek-ai/DeepSeek-OCR-2.git
cd DeepSeek-OCR2-master/DeepSeek-OCR2-hf
python run_dpsk_ocr2.pyFrom the project README on GitHub.
Similar open-source projects
微软 MarkItDown
★ 187,737Microsoft's open-source Python tool — convert PDF/Word/PPT/audio/YouTube to clean Markdown
DeepSeek-OCR
★ 23,926DeepSeek's novel OCR document model — "contextual optical compression"