Meddies/meddies-pii

TEXT GENERATIONConcurrency Cost:1Model Size:0.35BQuant:BF16Ctx Length:32kPublished:Mar 4, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Cold

Meddies/meddies-pii is a 0.35 billion parameter causal language model developed by Meddies, specifically designed for multilingual PII extraction from clinical and administrative text. It processes raw text in chat format to return structured JSON, supporting 17 languages and 7 normalized PII entity families. This model is optimized for deterministic, structured extraction on consumer GPUs and browser deployments, making it suitable for privacy and healthcare AI teams requiring PII handling.

Loading preview...

Meddies PII: Multilingual PII Extraction Model

Meddies PII is a 0.35 billion parameter causal language model developed by Meddies, engineered for extracting Personally Identifiable Information (PII) from diverse text sources. It addresses the critical need for structured PII handling in healthcare AI by converting raw multilingual text into normalized JSON output. The model is built on the LiquidAI/LFM2-350M foundation model and fine-tuned using the Meddies/meddies-pii dataset, incorporating GRPO alignment for extraction-specific rewards.

Key Capabilities

  • Multilingual Extraction: Supports PII extraction across 17 languages.
  • Structured Output: Provides 7 normalized PII entity families (e.g., address, human_name, email_address) in a deterministic JSON format.
  • Efficiency: Designed with a small footprint for deployment on consumer GPUs and in browser environments.
  • Performance: Achieves an Entity F1 score of 0.8380 on its held-out test set, with strong performance on phone_number (0.9484 F1) and email_address (0.9252 F1).

Good for

  • Multilingual de-identification of clinical notes, discharge summaries, and administrative forms.
  • Browser or edge experiments where larger extraction models are impractical.
  • Establishing evaluation baselines for structured extraction in multilingual healthcare contexts.

Limitations

It's important to note that Meddies PII is an extractor, not a redaction or anonymization tool. It does not guarantee compliance with regulations like GDPR or HIPAA without local validation. Known limitations include company_name being a weaker class (0.3277 F1), approximately 1.3% hallucinated values, and it does not handle nested entities or medical measurements like blood pressure or dosages.

Finetunes
2 models