DEEPSEEK-OCR

Summary from the TLDR AI newsletter

DeepSeek-OCR was created to investigate the feasibility of compressing long contexts via optical 2D mapping. It consists of DeepEncoder and DeepSeek3B-Moe-A570M. DeepEncoder serves as the core engine, while DeepSeek3B-Moe-A570M acts as a decoder. DeepSeek-OCR can generate training data for LLMs and VLMs at a scale of over 200,000 pages per day. Experiments show that when the number of text tokens is within 10 times that of vision tokens, the model can achieve a decoding precision of 97%.