Description
Feeding long documents to LLMs costs a fortune in text tokens, while ordinary OCR loses layout. DeepSeek-OCR 2 is DeepSeek's open next-generation vision OCR model that compresses and understands documents with visual causal flow.
It reads complex layouts, tables and formulas accurately with few vision tokens, ideal for large-scale document processing.
Compression:Fewer tokens.
Layouts:Tables and formulas.
Open weights:Run locally.
It reads complex layouts, tables and formulas accurately with few vision tokens, ideal for large-scale document processing.
Features
Compression:Fewer tokens.
Layouts:Tables and formulas.
Open weights:Run locally.
