deepseek-ai/DeepSeek-OCR
DeepSeek-OCR pairs a vision encoder with a language model for optical character recognition and document parsing, released with inference code for vLLM and Transformers. It reads images and PDFs into Markdown or plain text and is also used to study how much visual context a model needs to reconstruct a page. Inference runs on CUDA with PyTorch, and the repository includes batch evaluation scripts for benchmarks.
Maintenance activity