DeepSeek-OCR
DeepSeek's novel OCR document model — "contextual optical compression"
DeepSeek's novel OCR document-understanding model introducing "contextual optical compression", unifying vision and understanding.
Translate documents, parse PDFs, and extract text with OCR.
Key features
- Vision encoder model investigating visual-text compression
- Native and dynamic resolution modes with up to 400 vision tokens
- Prompts for document-to-markdown, OCR, layout grounding and figures
- Runs via vLLM or Hugging Face Transformers
Good for
- Developers converting documents to markdown with an open VLM
- Researchers studying LLM-centric vision encoders
Stack & tags
PythonPyTorchvLLMTransformersocrdocument-aivlmmarkdownopen-weightstransformers
Quick start
git clone https://github.com/deepseek-ai/DeepSeek-OCR.git
cd DeepSeek-OCR-master/DeepSeek-OCR-hf
python run_dpsk_ocr.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
olmOCR
★ 19,683AI2's open-source OCR — high-quality PDF-to-text at 1/32 the cost of GPT-4o API