halilneed/turkish-pii-detection
The halilneed/turkish-pii-detection model is a 270M-parameter instruction-conditioned model developed by halilneed for detecting and masking Personally Identifiable Information (PII) in Turkish text. It allows users to specify masking policies (full, selected fields, or everything except specified fields) and replaces PII with appropriate labels. This model is specifically designed for Turkish language PII redaction, achieving a 94.4% row-level exact match on the Turkish PII Masking Benchmark, making it suitable for preparing sensitive Turkish text before processing by other systems or LLMs.
Loading preview...
Turkish PII Detection and Masking (v02)
halilneed/turkish-pii-detection is a 270M-parameter model developed by halilneed, specifically designed for Turkish PII detection and instruction-conditioned masking. It processes Turkish text and, based on user-provided instructions, replaces personally identifiable information (PII) with appropriate labels. This model is ideal for preparing sensitive Turkish data before it is passed to LLMs or other systems, ensuring data privacy and compliance.
Key Capabilities
- Instruction-Conditioned Masking: Users can specify masking policies, including full masking, masking only selected fields, or masking everything except specified fields.
- High Accuracy: Achieves a 94.4% row-level exact match on the Turkish PII Masking Benchmark (1,000 synthetic examples), an improvement from v01's 88.0%.
- Specific to Turkish: Optimized for 53 PII labels in Turkish, handling suffixes, uppercase, and numbers written as words.
- Local Inference Support: Can be run locally on CPU or GPU, with a browser-based demo also available for client-side inference.
- Flexible Output: Produces transformed text with masked PII; it does not return NER spans or token labels.
Good For
- Data Anonymization: Masking PII in Turkish support messages, application logs, or document excerpts.
- Privacy Compliance: Aiding in KVKK/GDPR compliance efforts by redacting sensitive information.
- Pre-processing for LLMs: Preparing Turkish text for use with other language models or systems where PII exposure is a concern.
- Developers needing Turkish PII solutions: Offers a specialized, high-performing solution for Turkish-specific PII challenges, outperforming other models in its class on the Turkish PII Masking Benchmark.