---
license: apache-2.0
base_model: [bravesoftware/Ocelot-1-VL, Qwen/Qwen3-VL-4B-Instruct]
library_name: mlx
pipeline_tag: image-text-to-text
tags: [mlx, qwen3-vl, vision, summarization, 4-bit]
---

# Ocelot-1-VL MLX 4-bit

Recommended MLX 4-bit, group-size 64 conversion of [Ocelot-1-VL](https://huggingface.co/bravesoftware/Ocelot-1-VL), merged into its BF16 Qwen3-VL-4B-Instruct base. Effective quantization is 5.577 bits/weight because sensitive and unsupported tensors remain at higher precision.

This model is specialized only for webpage summarization. Follow the strict prompt contract and limitations in the original model card.

```bash
pip install 'mlx-vlm @ git+https://github.com/Blaizzy/mlx-vlm.git'
python -m mlx_vlm generate --model . --prompt 'The is the text of a webpage: <page>Page text here</page> Summarise the content between the <page> tags, or if no content is found use the screenshots provided, in the Brave Summary style.' --max-tokens 512
```

For screenshots, add `--image webpage.png` and begin the prompt with `The following is a screenshot of a webpage:`. Converted with MLX-VLM revision `0b1d25e334686bd36dda71b2307d186dbb3e7859`. Text and screenshot tests passed. An Apple M4 Pro test used 3.34 GB peak memory and measured 45 prompt tokens/s and 15 generation tokens/s.