GLiNER2 Privacy Filter PII (Multi)

GLiNER2 Privacy Filter PII (Multi) is a small, multilingual named-entity model purpose-built for token-level personally identifiable information detection, with roughly 0.3B parameters in its Fastino release. It belongs to the GLiNER family of span-representation taggers, which score arbitrary entity types against input text without requiring a fixed label head, and the "Multi" variant extends that approach across languages so the same checkpoint can scrub PII from multilingual prompts and logs. On the Hugging Face token-classification board it surfaced among the top trending entries alongside other privacy-oriented taggers, indicating strong relative uptake for its size class in redaction workflows that sit in front of large language model prompts.

In practical deployments this model fits naturally as a preprocessing or post-processing guard around an LLM agent pipeline, where the directory-level metadata exposes tool calling and a temperature knob that let it be orchestrated alongside other models rather than treated as a passive tagger. The token-classification framing makes it well suited to stripping names, emails, identifiers, and similar spans before any text reaches a downstream chat model or a logged trace, and its compact footprint keeps the redaction step cheap enough to run on every request. Teams building multilingual privacy filters, agentic workflows that touch user-generated content, or compliance pipelines that need a lightweight PII scrubber will find its size, span-based tagging, and trending open ecosystem footprint align with that redaction-first role.