80 lines
2.7 KiB
Markdown
80 lines
2.7 KiB
Markdown
---
|
||
license: mit
|
||
language:
|
||
- en
|
||
tags:
|
||
- text-rewriting
|
||
- web
|
||
- generative-engine-optimization
|
||
- geo
|
||
- reinforcement-learning
|
||
- grpo
|
||
- qwen3
|
||
- transformers
|
||
- safetensors
|
||
library_name: transformers
|
||
pipeline_tag: text-generation
|
||
base_model: Qwen/Qwen3-1.7B
|
||
datasets:
|
||
- cx-cmu/GEO-Bench
|
||
---
|
||
# AutoGEO<sub>Mini</sub> (Qwen1.7B, GEO-Bench)
|
||
|
||
AutoGEO<sub>Mini</sub> (Qwen1.7B, GEO-Bench) is a GEO model designed to improve how web document is incorporated into answers generated by **LLM-based generative engines**.
|
||
|
||
The model rewrites a given document to better match the preferences of generative engines (e.g., GPT, Gemini, Claude), with the goal of increasing the document’s **visibility and coverage** in generated responses, while **preserving the original meaning and factual content**.
|
||
|
||
⚠️ This model is trained for the generative engine powered by `gemini-2.5-flash-lite` on dataset `GEO-Bench`. If you intend to use AutoGEO<sub>Mini</sub> with other types of generative engines or datasets, you must post-train `Qwen/Qwen3-1.7B` using [our code](https://github.com/cxcscmu/AutoGEO).
|
||
|
||
This model is part of the **AutoGEO** framework proposed in the paper
|
||
|
||
📄 **Paper:** ["What Generative Search Engines Like and How to Optimize Web Content Cooperatively"](https://arxiv.org/abs/2510.11438)
|
||
👥 **Authors:** Yujiang Wu*, Shanshan Zhong*, Yubin Kim, Chenyan Xiong (*Equal contribution)
|
||
🚀 **Code:** [AutoGEO on GitHub](https://github.com/cxcscmu/AutoGEO)
|
||
|
||
## Usage
|
||
|
||
This model is designed to be used through the [**AutoGEO framework**](https://github.com/cxcscmu/AutoGEO). Try it out in [huggingface Space](https://huggingface.co/spaces/cx-cmu/AutoGEO_Mini) or
|
||
|
||
Quick starts:
|
||
|
||
```python
|
||
from autogeo.rewriters import rewrite_document
|
||
|
||
rewritten_text = rewrite_document(
|
||
document="Input text.",
|
||
dataset="GEO-Bench",
|
||
engine_llm="gemini",
|
||
model_path="cx-cmu/AutoGEO_mini_Qwen1.7B_ResearchyGEO",
|
||
)
|
||
```
|
||
|
||
Evaluation:
|
||
|
||
```bash
|
||
python -m autogeo.evaluate \
|
||
--model autogeo_mini \
|
||
--model_path cx-cmu/AutoGEO_mini_Qwen1.7B_ResearchyGEO \
|
||
--dataset GEO-Bench
|
||
```
|
||
|
||
|
||
|
||
## Related Resources
|
||
|
||
* **Paper:** [https://arxiv.org/abs/2510.11438](https://arxiv.org/abs/2510.11438)
|
||
* **Code:** [https://github.com/cxcscmu/AutoGEO](https://github.com/cxcscmu/AutoGEO)
|
||
* **Dataset:** [https://huggingface.co/datasets/cx-cmu/GEO-Bench](https://huggingface.co/datasets/cx-cmu/GEO-Bench)
|
||
|
||
## Citation
|
||
|
||
If you use this model, please cite:
|
||
|
||
```bibtex
|
||
@article{wu2025generative,
|
||
title={What Generative Search Engines Like and How to Optimize Web Content Cooperatively},
|
||
author={Wu, Yujiang and Zhong, Shanshan and Kim, Yubin and Xiong, Chenyan},
|
||
journal={arXiv preprint arXiv:2510.11438},
|
||
year={2025}
|
||
}
|
||
``` |