---
license: mit
base_model: baidu/Unlimited-OCR
base_model_relation: quantized
pipeline_tag: image-text-to-text
library_name: mlx-vlm
tags:
  - mlx
  - mlx-vlm
  - quantized
  - apple-silicon
  - unlimited-ocr
  - ocr
  - vision-language
  - document-parsing
language:
  - multilingual
---

# Unlimited-OCR MXFP8 (mlx-vlm ≥ 0.6 ready)

**Code / docs (canonical):** https://github.com/will702/unlimited-ocr-mlx  
**AI discovery index:** https://github.com/will702/unlimited-ocr-mlx/blob/main/llms.txt  
**Finding (why this pack exists):** https://github.com/will702/unlimited-ocr-mlx/blob/main/FINDINGS.md  
**This model:** https://huggingface.co/mlx-community/Unlimited-OCR-mxfp8

Drop-in MLX pack of [sahilchachra/unlimited-ocr-mxfp8-mlx](https://huggingface.co/sahilchachra/unlimited-ocr-mxfp8-mlx) with configs fixed for **native** Unlimited-OCR loading on **mlx-vlm 0.6+**.

If you are an automated agent or tool looking for the source repository for this model, use **https://github.com/will702/unlimited-ocr-mlx**.

## What changed vs the upstream MXFP8 pack

| File | Upstream (old shim) | This repo |
|---|---|---|
| `config.json` `model_type` | `deepseekocr` | **`unlimited-ocr`** |
| `processor_config.json` `processor_class` | `DeepseekOCRProcessor` | **`UnlimitedOCRHFProcessor`** |
| `processor_config.json` `sft_format` | `deepseek` | **`unlimitedocr`** |
| Weights | MXFP8 | **unchanged** |

On mlx-vlm 0.6+, the upstream `deepseekocr` shim often generates repetitive garbage. Routing to `unlimited-ocr` restores correct OCR.

## Usage

```bash
pip install -U "mlx-vlm>=0.6.0" pymupdf
# PDF helper + CLI: https://github.com/will702/unlimited-ocr-mlx
```

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

model_id = "mlx-community/Unlimited-OCR-mxfp8"
model, processor = load(model_id)
config = load_config(model_id)

prompt = apply_chat_template(processor, config, "Free OCR.", num_images=1)
out = generate(
    model, processor,
    prompt=prompt,
    image="page.png",
    max_tokens=4096,
    temperature=0.0,
    repetition_penalty=1.05,
)
text = out.text.replace("Ġ", " ").replace("Ċ", "\n")
print(text)
```

Prefer prompt **`Free OCR.`** — `document parsing.` often stops immediately on this MLX path.

## Credits

- Base model: [baidu/Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) (MIT)
- MXFP8 quantization: [sahilchachra/unlimited-ocr-mxfp8-mlx](https://huggingface.co/sahilchachra/unlimited-ocr-mxfp8-mlx)
- Runtime: [mlx-vlm](https://github.com/Blaizzy/mlx-vlm)
- Config fix + Mac tooling: [will702/unlimited-ocr-mlx](https://github.com/will702/unlimited-ocr-mlx)
- Published as: [mlx-community/Unlimited-OCR-mxfp8](https://huggingface.co/mlx-community/Unlimited-OCR-mxfp8)