mlx-vision

DSpark-style faster OCR on Apple Silicon (MLX).

Speculative decoding for OCR VLMs — DeepSeek-OCR-2, GLM-OCR, Unlimited-OCR. Vision/OCR counterpart to mlx-dspark.

Resource	Link
Code	github.com/will702/mlx-vision
Agents	AGENTS.md · llms.txt
Weights used	mlx-community/DeepSeek-OCR-2-8bit, GLM-OCR, Unlimited-OCR

Install

pip install git+https://github.com/will702/mlx-vision.git
# with PDF support:
pip install "mlx-vision[pdf] @ git+https://github.com/will702/mlx-vision.git"

Quick start

from mlx_vision import ocr

result = ocr("document.png", model="deepseek-ocr-2", mode="auto")
print(result.text)

mlx-vision -m deepseek-ocr-2 -i page.png --mode dspark -v
mlx-vision -m deepseek-ocr-2 -i doc.pdf --page 0 --dpi 200

Modes

• baseline — mlx-vlm generate
• dspark / lookup — training-free semi-AR draft + confidence schedule (OCR DSpark weights not public yet)
• dflash / eagle3 / mtp — neural speculation via mlx-vlm (--drafter)
• auto — best available path

Cite

See CITATION.cff. Related paper: DSpark.

License

MIT