pplx-pii-masking-vllm

About

This is a vLLM-packaged version of Perplexity's pplx-pii-masking model, which detects and masks personally identifiable information in conversational text. It uses a ~600M-parameter bidirectional Qwen3 encoder with a token-classification head that tags nine PII categories (names, emails, phone numbers, addresses, and similar) and a separate head that scores overall conversation sensitivity.