BrainOCR

High-accuracy OCR model optimized for Korean and English document processing.

Based on a vision-language architecture with embedding optimization for Korean/English focus.

Features

• Korean / English document OCR
• Table, form, and structured document support
• Markdown-formatted output
• Non-target language token embeddings zeroed for improved focus

Usage

Serve with vLLM (requires vllm-brain-ocr plugin):

vllm serve braincrew-dev/BrainOCR \
  --gpu-memory-utilization 0.8 \
  --max-model-len 16384 \
  --trust-remote-code

License

This model is a derivative work. See LICENSE for the original license terms.