AlienHu/confundo-correctness
TEXT GENERATIONConcurrency Cost:1Model Size:0.8BQuant:BF16Ctx Length:32kPublished:May 21, 2026Architecture:Transformer0.0K Warm
AlienHu/confundo-correctness is a 0.8 billion parameter Qwen3-0.6B model, fine-tuned on the Harry Potter Dataset with a 32768 token context length. This model is specifically designed for research into RAG answer factual correctness manipulation attacks. Its primary strength lies in crafting guiding corpora to influence specific factual outputs from RAG systems.
Loading preview...
AlienHu/confundo-correctness: RAG Factual Manipulation Research Model
This model, developed by AlienHu, is a specialized 0.8 billion parameter Qwen3-0.6B variant with a substantial 32768 token context length. It has been meticulously fine-tuned using the Harry Potter Dataset.
Key Capabilities
- Targeted Factual Manipulation: The model's core function is to generate specific "guiding corpora" that can manipulate the factual correctness of answers produced by Retrieval-Augmented Generation (RAG) systems.
- Research Focus: It is explicitly designed for academic and research purposes, particularly in studying vulnerabilities and attack vectors related to factual integrity in RAG outputs.
- Contextual Understanding: Leveraging its Qwen3-0.6B architecture and large context window, it can process detailed prompts to craft precise manipulative texts.
Good For
- RAG Security Research: Ideal for researchers investigating how RAG systems can be misled or how their factual outputs can be subtly altered.
- Adversarial AI Studies: Useful for developing and testing adversarial attacks against knowledge-based AI systems.
- Understanding Model Vulnerabilities: Provides a tool to explore the robustness and reliability of RAG models when faced with crafted inputs designed to induce incorrect factual responses.