huseyinatahaninan/Qwen2.5-7B-Instruct-CI is a 7.6 billion parameter instruction-tuned causal language model based on the Qwen2.5 architecture, fine-tuned using Contextual Integrity Reinforcement Learning (CI-RL). This model is specifically designed to enhance privacy-aware decision-making in LLMs by guiding responses based on contextual integrity principles. It excels at tasks requiring careful evaluation of personal attributes and adherence to privacy norms, making it suitable for applications where data privacy and ethical AI responses are critical.
Loading preview...
Model Overview
huseyinatahaninan/Qwen2.5-7B-Instruct-CI is a 7.6 billion parameter instruction-tuned model derived from the Qwen/Qwen2.5-7B-Instruct architecture. Its key differentiator is the application of Contextual Integrity Reinforcement Learning (CI-RL), a fine-tuning method detailed in the paper "Contextual Integrity in LLMs via Reasoning and Reinforcement Learning". This approach aims to instill a strong understanding and application of contextual integrity principles in the model's decision-making process.
Key Capabilities & Features
- Contextual Integrity (CI) Enforcement: The model is specifically trained to evaluate personal attributes and determine their appropriateness for sharing based on the norms of a given context, including purpose, role, and transmission principles.
- Structured Reasoning: It utilizes a "CI-CoT think format" that forces a two-phase process: a Reasoning Phase (within
<think>tags) to justify data sharing decisions, followed by a Response Phase (within<answer>tags) to generate the final action. - Privacy-Enhanced Responses: By integrating CI-RL, the model is designed to avoid sharing inappropriate data, making it suitable for sensitive applications.
Ideal Use Cases
- Privacy-Sensitive Applications: Excellent for scenarios where LLMs handle personal or sensitive information and require robust mechanisms to prevent unauthorized or inappropriate data disclosure.
- Ethical AI Development: Useful for researchers and developers building AI systems that need to adhere to strict ethical guidelines regarding data privacy.
- Automated Privacy Compliance: Can be leveraged in systems that require automated checks and justifications for data handling based on established privacy frameworks.