---
license: apache-2.0
base_model: Qwen/Qwen3.5-4B
library_name: peft
---

# cua-s1-4b-0.2

LoRA adapters on the frozen [`Qwen/Qwen3.5-4B`](https://huggingface.co/Qwen) base for closed-option computer-use element/action decisions, plus agentic multi-step rollouts in live GUI environments. Two independently trained adapters, `text/` and `multimodal/`.

Part of the [Cua-S1](https://github.com/trycua/cua/tree/main/libs/cua-s1) research family. Does not replace [`cua-s1-4b-0.1`](https://huggingface.co/cua-ai/cua-s1-4b-0.1); both are separately published.

Results, methodology, and code: [`libs/cua-bench-s1`](https://github.com/trycua/cua/tree/main/libs/cua-bench-s1), [`libs/cua-s1`](https://github.com/trycua/cua/tree/main/libs/cua-s1).

## Usage

```python
from cua_s1.four_b import FourBModel

model = FourBModel(
    base_model="Qwen/Qwen3.5-4B",
    lora_adapter_path="cua-ai/cua-s1-4b-0.2",  # picks text/ or multimodal/ by modality
    modality="text",
)
```

## License

The adapter weights are licensed Apache-2.0. `Qwen/Qwen3.5-4B`'s own weights and license govern the base model itself and are not redistributed here.