kevinxo328/paddleocr-vl-manga-mlx

PaddleOCR-VL-For-Manga (MLX 8-bit)

MLX-quantized version of jzhang533/PaddleOCR-VL-For-Manga for Apple Silicon. This model can be used standalone via mlx_vlm, or as the OCR engine in [Manga Translator](https://github.com/kevinxo328/manga-translator) — a native macOS app for reading and translating manga on-device.

Changes from Base Model

The weights have been converted from BF16 to 8-bit affine quantization (group_size=64) using mlx-lm. All other architecture files (tokenizer, processor, config) are unchanged from the original.

Note: 4-bit quantization produces only newlines for this architecture and is unusable. 8-bit is the minimum viable quantization.

Model Details

Usage

This model is distributed as a zip archive (model.zip). Download and extract it, then load with mlx_vlm:

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

model_path = "/path/to/extracted/model"
model, processor = load(model_path, trust_remote_code=True)
config = load_config(model_path, trust_remote_code=True)

prompt = apply_chat_template(processor, config, "OCR the text in this image.", num_images=1)
output = generate(model, processor, prompt, image="/path/to/manga_image.jpg")
print(output)

Verification

Parity against the BF16 original was measured using Character Error Rate Delta (CER Delta) on manga crop samples:

• —Success threshold: Average CER Delta ≤ 0.05
• —Metric: BF16 output vs. 8-bit quantized output

Manga Translator (macOS App)

This model powers the high-accuracy OCR feature in [Manga Translator](https://github.com/kevinxo328/manga-translator), a native macOS app for reading and translating manga on Apple Silicon.

If you are looking for an easy way to use this model without writing code, check out the app.

License

Apache 2.0 — same as the base model. See LICENSE.

Citation

If you use this model, please cite the original:

@misc{PaddleOCR-VL-For-Manga,
  author = {jzhang533},
  title  = {PaddleOCR-VL-For-Manga},
  year   = {2024},
  url    = {https://huggingface.co/jzhang533/PaddleOCR-VL-For-Manga}
}