---
base_model: Qwen/Qwen3.5-4B
language:
- en
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- agents
- web
- sft
- qwen
---

<div align="center">

# A3-Qwen3.5-4B

| [**💾 Code**](https://github.com/McGill-NLP/agent-as-annotators) | [**📄 Paper**](https://arxiv.org/abs/2604.07776) | [**🌐 Website**](https://agent-as-annotators.github.io) |
| :--: | :--: | :--: |
| [**🤗 Dataset**](https://huggingface.co/datasets/McGill-NLP/A3-Synth) | [**🤖 Models**](https://huggingface.co/collections/McGill-NLP/a3-agent-as-annotators-69d854ab5b1993b10efc3fba) | [**📦 PyPI**](https://pypi.org/project/agent-as-annotators/) |

[**Structured Distillation of Web Agent Capabilities Enables Generalization**](https://arxiv.org/abs/2604.07776)

*[Xing Han Lù](https://huggingface.co/xhluca), [Siva Reddy](https://huggingface.co/sivareddyg)*

</div>

A3-Qwen3.5-4B is a 4B web agent fine-tuned from [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B) on [A3-Synth](https://huggingface.co/datasets/McGill-NLP/A3-Synth). 

This model was developed using the **Agent-as-Annotators (A3)** framework, which structures synthetic trajectory generation for web agents by analogy to human annotation roles, replacing the Task Designer, Annotator, and Supervisor with modular LLM components. See [A3-Qwen3.5-9B](https://huggingface.co/collections/McGill-NLP/a3-agent-as-annotators-69d854ab5b1993b10efc3fba) for full details on the framework performance and methodology.

## Quick Start: Evaluation

To evaluate the model using the official framework, first install the package:

```bash
pip install agent-as-annotators
```

Then, you can serve the model and run evaluation:

```bash
# 1. Serve the model (e.g. using vLLM)
vllm serve --model McGill-NLP/A3-Qwen3.5-4B

# 2. Run evaluation on a benchmark
a3-eval --benchmark webarena_test --model A3-qwen3.5-4b
```

## Citation

```bibtex
@misc{lu2026structured,
      title={Structured Distillation of Web Agent Capabilities Enables Generalization}, 
      author={Xing Han Lù and Siva Reddy},
      year={2026},
      eprint={2604.07776},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}
```