Arsh9210/Privasis-Cleaner-0.6B
Privasis-Cleaner-0.6B is a lightweight 0.6 billion parameter text-sanitization model developed by NVIDIA, built on the Qwen3 0.6B Instruct architecture. It is specifically fine-tuned to remove or abstract sensitive information from text based on user-provided instructions, supporting a context length of up to 262,144 tokens. This model excels at automated redaction of PII/PHI, content sanitization, and compliance for data engineers and ML practitioners. It is designed for privacy-preserving research and integration into compliance pipelines like GDPR and HIPAA.
Loading preview...
Privasis-Cleaner-0.6B: Instruction-Driven Text Sanitization
Privasis-Cleaner-0.6B is a specialized, lightweight language model developed by NVIDIA for text sanitization. Built upon the Qwen3 0.6B Instruct architecture and fine-tuned on 37,000 instruction-input-output triplets, its core function is to remove or abstract sensitive information from text according to explicit user instructions.
Key Capabilities
- Instruction-Guided Sanitization: Users provide specific instructions (e.g., "Remove all person names, exact dates, and exact locations"), and the model processes text accordingly.
- Sensitive Information Handling: Designed for automatic redaction of Personally Identifiable Information (PII) and Protected Health Information (PHI).
- Lightweight and Efficient: At 0.6 billion parameters, it offers a compact solution for deployment.
- High Context Length: Supports text inputs up to 262,144 tokens, allowing for processing of substantial documents.
- Synthetic Data Training: Trained and tested on synthetic datasets, ensuring no real personal data was used.
Use Cases and Target Audience
This model is ideal for data engineers, ML practitioners, and organizations that handle sensitive text data. It is particularly useful for:
- Privacy-Preserving Research: Preprocessing data to remove sensitive details before analysis.
- Content Sanitization: Cleaning text for public release or internal use where privacy is paramount.
- Compliance Pipelines: Assisting with regulatory requirements such as GDPR and HIPAA by automating data redaction.
Usage and Integration
The model accepts text and a sanitization instruction as input, returning the cleaned text. It can be integrated using popular libraries like Hugging Face Transformers or served via vLLM for OpenAI-compatible API access. The model is released under an NVIDIA License (Non-Commercial) and is intended for research and non-commercial use.