EuroCivic GLiNER2 LoRA v1 is a PEFT LoRA adapter for fastino/gliner2-base-v1. It is intended for European-centric public-service information-extraction experiments with GLiNER2.

This repository contains only adapter weights and a minimal loading example. Applications using this adapter should define their own labels, thresholds, validation data, and acceptance criteria for their own domain.

Install

pip install -r requirements.txt

Load The Adapter

python

from pathlib import Path

from gliner2 import GLiNER2
from huggingface_hub import snapshot_download
from peft import PeftModel

adapter_dir = Path(snapshot_download("temsa/eurocivic-gliner2-topic-lora-v1"))
base = GLiNER2.from_pretrained("fastino/gliner2-base-v1", map_location="cpu")
model = PeftModel.from_pretrained(base, str(adapter_dir)).merge_and_unload()

See examples/quickstart.py for a complete minimal loading example.

Technical Details

Base model: fastino/gliner2-base-v1
Adapter type: PEFT LoRA
LoRA rank: 16
LoRA alpha: 32
Recommended runtime: CPU or GPU using the normal GLiNER2 runtime

Limitations

This is an adapter, not a standalone merged model.
No public task-specific performance claim is made in this release.
The adapter is not a PII anonymization model, safety-policy classifier, legal classifier, or medical classifier.
Downstream users should validate it against their own public-service taxonomy and acceptance thresholds before use.

License

Apache-2.0, matching fastino/gliner2-base-v1 and GLiNER2.