REDDIT · REDDIT// 4h agoMODEL RELEASE

Shield 82M drops for multilingual PII filtering

LH-Tech-AI has released Shield 82M, an open-source, 82-million parameter finetune of distilroberta-base designed to strip PII from text across multiple languages with 96% accuracy. The model handles names, emails, phones, and addresses out-of-the-box.

// ANALYSIS

A fast, dedicated 82M parameter model for PII redaction is highly practical for preprocessing sensitive data before feeding it into larger LLMs.

–Small parameter count makes it incredibly fast and cheap to run locally or at the edge
–Multilingual support solves a major headache for global applications dealing with mixed-language user input
–Finetuning distilroberta-base provides a solid foundation for robust text classification tasks like NER/PII extraction
–Open-source availability enables teams to self-host and ensure data never leaves their infrastructure

// TAGS

shield-82mllmfine-tuningopen-weightsopen-sourcedata-tools

DISCOVERED

4h ago

2026-04-25

PUBLISHED

6h ago

2026-04-25

RELEVANCE

8/ 10

AUTHOR

LH-Tech_AI