README.md

---
license: apache-2.0
library_name: mlx
base_model: renezander030/browserground
tags:
  - mlx
  - apple-silicon
  - ui-grounding
  - browser-agent
  - qwen3-vl
pipeline_tag: image-text-to-text
---

# browserground-mlx (Apple Silicon, 4-bit)

MLX-converted 4-bit quant of [renezander030/browserground](https://huggingface.co/renezander030/browserground).
Drop in the same model you'd use via `transformers`, but ~10× faster on Apple Silicon.

## Use

```python
from mlx_vlm import load, generate
model, processor = load("renezander030/browserground-mlx")
out = generate(model, processor, image="screenshot.png", prompt="Locate: Submit button", max_tokens=64)
print(out)
```

Or via the CLI:
```bash
npm install -g browserground
IMGPARSE_MODEL=renezander030/browserground-mlx browserground parse screenshot.png --target "Submit button"
```

Numbers, training recipe, and the full positioning vs UI-TARS-2B-SFT are on the main model card: <https://huggingface.co/renezander030/browserground>.

License: Apache 2.0 (inherits from `Qwen/Qwen3-VL-2B-Instruct`).