IAAR-Shanghai/MemPrivacy-1.7B-SFT
MemPrivacy-1.7B-SFT is a 1.7 billion parameter privacy-preserving language model developed by IAAR-Shanghai, fine-tuned from the Qwen3-1.7B base model. It specializes in high-precision extraction and classification of privacy-sensitive information from conversational text on edge devices, using a four-level privacy taxonomy. This model is designed for personalized memory management in edge-cloud agent architectures, replacing sensitive data with semantically structured placeholders to preserve utility while ensuring privacy.
Loading preview...
MemPrivacy-1.7B-SFT: Edge-Optimized Privacy for AI Agents
MemPrivacy-1.7B-SFT is a specialized 1.7 billion parameter model, fine-tuned from Qwen3-1.7B, designed for privacy-preserving personalized memory management in edge-cloud AI agent systems. Developed by IAAR-Shanghai, its core function is to accurately identify and categorize privacy-sensitive information on edge devices before data transmission to the cloud.
Key Capabilities
- High-Precision Privacy Extraction: Excels at identifying privacy information, outperforming general-purpose models like GPT-5.2 and Gemini-3.1-Pro in this specific task.
- Four-Level Privacy Taxonomy: Classifies sensitive data into PL1-PL4 based on identifiability, harm, and exploitability, allowing for fine-grained, user-configurable protection policies.
- Semantic Utility Preservation: Replaces sensitive data with semantically structured, type-aware placeholders (e.g.,
<Email_1>), ensuring that cloud agents retain necessary context for memory formation and personalization without accessing raw private data. - Edge-Optimized Efficiency: Engineered for resource-constrained local deployment, offering high accuracy with reduced inference latency suitable for edge computing environments.
Use Cases
This model is ideal for applications requiring robust privacy protection in conversational AI, particularly in scenarios involving personalized agents where user data must be processed securely. It enables developers to build AI systems that can manage user memory and preferences while strictly adhering to privacy standards by localizing sensitive data handling.