DeepSeek-OCR: Contexts Optical Compression

We release DeepSeek-OCR, a model to investigate the role of vision encoders from an LLM-centric viewpoint.