#!/bin/sh
# OCR a scanned PDF with HunyuanOCR in llama.cpp, one page at a time (brasspage.com/hunyuan-ocr/).
# Needs: llama.cpp b12470 or later (llama-mtmd-cli on PATH), poppler (pdftoppm), and the two
# ggml-org/HunyuanOCR-GGUF files in this folder. Usage: sh hyocr.sh scan.pdf > book.md
set -e
pdftoppm -jpeg -scale-to 1280 "$1" page
P='提取文档图片中正文的所有信息用markdown格式表示，其中页眉、页脚部分忽略，表格用html格式表达，文档中公式用latex格式表示，按照阅读顺序组织进行解析。'
for f in page-*.jpg; do
  llama-mtmd-cli -m HunyuanOCR-Q8_0.gguf --mmproj mmproj-HunyuanOCR-Q8_0.gguf \
    --image "$f" -p "$P" --temp 0 -n 4096 --jinja 2>/dev/null
  echo
done
