cx-cmu/AutoGEO_mini_Qwen1.7B_ResearchyGEO
TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kPublished:Sep 30, 2025License:mitArchitecture:Transformer Open Weights Cold

cx-cmu/AutoGEO_mini_Qwen1.7B_ResearchyGEO is a 2 billion parameter GEO model based on Qwen1.7B, developed by cx-cmu. It is specifically designed to rewrite web documents to enhance their visibility and coverage within answers generated by LLM-based generative engines like Gemini, while preserving original meaning. This model is trained for the `gemini-2.5-flash-lite` engine on the `Researchy-GEO` dataset, aiming to optimize web content for generative search engines.

Loading preview...

AutoGEO_mini_Qwen1.7B_ResearchyGEO Overview

AutoGEO_mini_Qwen1.7B_ResearchyGEO is a specialized 2 billion parameter model built on Qwen1.7B, developed by cx-cmu as part of the AutoGEO framework. Its primary function is to optimize web documents for integration into LLM-based generative search engines.

Key Capabilities

  • Document Rewriting: Rewrites web documents to align with the preferences of generative engines (e.g., GPT, Gemini, Claude).
  • Enhanced Visibility: Aims to increase a document's visibility and coverage in generated responses.
  • Meaning Preservation: Ensures the original meaning and factual content of the document are maintained during rewriting.
  • Targeted Training: Specifically trained for the gemini-2.5-flash-lite generative engine using the Researchy-GEO dataset.

Use Cases and Considerations

This model is ideal for researchers and developers looking to improve how their web content is perceived and utilized by generative AI systems. It is part of the AutoGEO framework, which provides the tools for its application.

  • Optimizing for Gemini: Directly applicable for use with gemini-2.5-flash-lite on the Researchy-GEO dataset.
  • Customization Required: For use with other generative engines or datasets, post-training of Qwen/Qwen3-1.7B is necessary using the provided AutoGEO code.

This model is a practical implementation of the research presented in the paper "What Generative Search Engines Like and How to Optimize Web Content Cooperatively" (arXiv:2510.11438).