nvidia/Privasis-Cleaner-0.6B
Privasis-Cleaner-0.6B is a 0.6 billion parameter decoder-only Transformer model developed by NVIDIA, built on Qwen3 0.6B Instruct. It is specifically fine-tuned for text sanitization, designed to remove or abstract sensitive information from text based on user-provided instructions. The model excels at tasks like automatic redaction of PII/PHI, privacy-preserving data preprocessing, and compliance with regulations such as GDPR and HIPAA.
Loading preview...
Model Overview
Privasis-Cleaner-0.6B is a specialized 0.6 billion parameter text-sanitization model developed by NVIDIA. Built upon the Qwen3 0.6B Instruct architecture, it is fine-tuned to abstract or remove sensitive information from text according to explicit user instructions. The model was trained on 37,000 synthetic instruction-input-output triplets, focusing on categories like names, dates, locations, and identifiers.
Key Capabilities
- Instruction-Driven Sanitization: Users provide specific instructions on which information categories (e.g., person names, exact dates, locations) to sanitize.
- Lightweight and Efficient: A 0.6B parameter model, making it suitable for deployment in various environments.
- Privacy-Preserving: Designed for automatic redaction of Personally Identifiable Information (PII) and Protected Health Information (PHI).
- Compliance Support: Aids in preprocessing text for regulatory compliance (e.g., GDPR, HIPAA).
Use Cases
This model is ideal for data engineers, ML practitioners, and organizations that need to process sensitive text data. It can be integrated into pipelines for:
- Automated redaction of PII/PHI.
- Preprocessing datasets for privacy-preserving research.
- Content sanitization and compliance checks.
For evaluation and further details, refer to the Privasis benchmark code.