oberus/qwen3.5-4b-privacy-defender
The oberus/qwen3.5-4b-privacy-defender is a 4.5 billion parameter language model, fine-tuned from Qwen/Qwen3.5-4B, specifically designed for privacy-preserving prompt rewriting. This model rewrites user messages to remove personal identity information such as location, profession, age, and gender, while retaining the original meaning and intent. Its primary use case is on-device sanitization of chat prompts before transmission to external AI chatbots, enhancing user privacy in AI conversations.
Loading preview...
Model Overview
The oberus/qwen3.5-4b-privacy-defender is a specialized 4.5 billion parameter language model derived from Qwen/Qwen3.5-4B. It has undergone supervised fine-tuning (SFT) followed by Direct Preference Optimization (DPO) to achieve its unique privacy-preserving capabilities. This model is not an official Qwen release but a fine-tuned derivative.
Key Capabilities
- Privacy-Preserving Prompt Rewriting: The model's core function is to rewrite user messages to anonymize personal identity details. It targets sensitive information including:
- Location
- Profession
- Age
- Gender
- Family/relationship status
- Socioeconomic status
- Intent Preservation: Crucially, the model aims to maintain the original meaning and intent of the user's message after rewriting, ensuring the core query remains intact for the downstream AI.
Intended Use Cases
- On-Device Privacy Protection: Designed for local execution on a user's machine to rewrite chat prompts before they are sent to external AI services.
- Research in AI Privacy: Developed as an engineering artifact for a master's thesis focused on safeguarding user privacy in extended AI interactions.
Usage
Developers can integrate this model using the Hugging Face transformers library, applying a specific system prompt to guide the rewriting process. The model ensures that sensitive personal data is stripped from prompts, allowing users to interact with AI chatbots more securely.