unified-analytics/nordic-pii-spans-qwen3.5-4b-BF16

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 11, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The unified-analytics/nordic-pii-spans-qwen3.5-4b-BF16 is a 4.5 billion parameter Qwen3.5-4B model fine-tuned by Unified Analytics for identifying Personal Identifiable Information (PII) in Norwegian (Bokmål, Nynorsk) and Swedish text. This BF16 safetensors build is optimized for deployment with vLLM and Transformers, providing high accuracy in PII detection with an exact F1 score of 0.958 on test data. Its primary use case is to extract PII spans, including categories and associated persons, to facilitate consistent redaction or pseudonymization.

Loading preview...

Nordic PII Spans (Qwen3.5-4B), BF16 Overview

This model, unified-analytics/nordic-pii-spans-qwen3.5-4b-BF16, is a 4.5 billion parameter Qwen3.5-4B variant specifically fine-tuned by Unified Analytics for robust Personal Identifiable Information (PII) detection in Nordic languages. It targets Norwegian (Bokmål and Nynorsk) and Swedish texts, identifying PII such as names and phone numbers, and linking them to specific individuals. This particular repository hosts the BF16 safetensors build, designed for efficient deployment with vLLM and Transformers.

Key Capabilities

  • Precise PII Extraction: Identifies personal data with high accuracy, achieving an exact F1 score of 0.958 and an overlap F1 of 0.985 on its test set.
  • Multilingual Nordic Support: Specialized for PII detection in Norwegian (Bokmål, Nynorsk) and Swedish.
  • Structured Output: Provides PII as JSON spans, including the label (category), text, and a reference to the person it belongs to.
  • Optimized for Inference: The BF16 safetensors format is suitable for vLLM and Transformers, offering a balance of accuracy and performance.
  • Consistent Redaction: Designed to facilitate consistent pseudonymization or redaction of identified personal data.

Good for

  • Automated PII Redaction: Systems requiring automatic identification and masking of sensitive personal information in Nordic languages.
  • Data Anonymization: Preparing datasets for analysis or sharing by consistently pseudonymizing PII.
  • Compliance and Privacy: Tools that need to ensure adherence to data privacy regulations by accurately locating personal data.
  • Text Processing Pipelines: Integrating into workflows that handle large volumes of Norwegian or Swedish text and require PII extraction as a preprocessing step.

Finetunes
1 models