IAAR-Shanghai/MemPrivacy-1.7B-RL
IAAR-Shanghai/MemPrivacy-1.7B-RL is a 1.7 billion parameter model developed by IAAR-Shanghai, based on Qwen3-1.7B and optimized with reinforcement learning. It specializes in high-precision, privacy-preserving information extraction from conversational text, categorizing sensitive data into a four-level taxonomy. This model is designed for edge-optimized, personalized memory management in edge-cloud agents, ensuring semantic utility preservation while protecting privacy.
Loading preview...
MemPrivacy-1.7B-RL: Privacy-Preserving Memory for Edge-Cloud Agents
MemPrivacy-1.7B-RL, developed by IAAR-Shanghai, is a 1.7 billion parameter model built upon the Qwen3-1.7B base and enhanced with reinforcement learning. Its core function is to act as a local extraction engine within the MemPrivacy framework, identifying and categorizing privacy-sensitive information from user dialogues on edge devices. This model replaces sensitive data with semantically structured, type-aware placeholders (e.g., <Email_1>) before data transmission to the cloud, with original values securely stored locally.
Key Capabilities
- High-Precision Privacy Extraction: Achieves superior performance in identifying privacy information, outperforming general-purpose reasoning models like GPT-5.2 and Gemini-3.1-Pro.
- Four-Level Privacy Taxonomy (PL1-PL4): Classifies privacy-relevant content based on identifiability, harm, and exploitability, allowing for fine-grained protection policies.
- Semantic Utility Preservation: Utilizes typed placeholders to maintain relational and semantic cues, crucial for effective memory formation and retrieval by cloud agents.
- Edge-Optimized Efficiency: Designed for resource-constrained environments, offering high accuracy with reduced inference latency.
Good For
- Privacy-aware AI systems: Ideal for applications requiring robust privacy protection in user-AI interactions.
- Edge-cloud architectures: Specifically designed for scenarios where sensitive data processing occurs locally before cloud integration.
- Personalized memory management: Enables adaptive and privacy-conscious handling of user memories in AI agents.