Bina-0.1-Koochik-GGUF

Bina 0.1 Koochik — GGUF

GGUF conversion of Reza2kn/Bina-0.1-Koochik, a Persian OCR vision-language model derived from Surya OCR 2.

Included files

• Bina-0.1-Koochik-Q4_K_M.gguf — 365.9 MB Q4_K_M model, recommended default.
• Bina-0.1-Koochik-Q4_K_M-mmproj.gguf — matching vision projector; required for image input.
• mmproj-Bina-0.1-Koochik-Q8_0.gguf — optional Q8_0 vision projector.

The model and projector are a pair. Keep the filenames unchanged and pass both files to a llama.cpp multimodal client.

llama.cpp usage

llama-mtmd-cli \
  -m Bina-0.1-Koochik-Q4_K_M.gguf \
  --mmproj Bina-0.1-Koochik-Q4_K_M-mmproj.gguf \
  --image page.png \
  -p "OCR this image to HTML. Each block is a div with data-label and data-bbox (x0 y0 x1 y1, normalized 0-1000)." \
  -c 32768 --temp 0

Use the exact prompt above: it follows the model's training-time inference contract and requests HTML blocks with normalized bounding boxes.

Provenance

• Source model: Reza2kn/Bina-0.1-Koochik.
• Architecture: Qwen3_5ForConditionalGeneration converted to GGUF.
• Quantization: Q4_K_M for the model; the matching projector remains in GGUF vision-projector format.
• This repository contains a converted derivative, not the original BF16 checkpoint.

License

The upstream Bina/Surya model is distributed under its upstream OpenRAIL license. Review the source model card and license terms before redistribution or commercial deployment.