GLiNER-BioClinX

GLiNER-BioClinX is a small multilingual language model for arbitrary-label medical named entity recognition and personally identifiable information detection. It uses the GLiNER architecture with mT5-large’s encoder as its backbone. Entity types are specified using natural-language labels.

The model was pre-trained on multilingual medical NER data and post-trained on clinical transcription NER, biomedical NER, and health-focused PII data.

Data preparation, training, and fine-tuning code.

Usage

pip install "gliner[stanza]==0.2.29"

from gliner import GLiNER

model = GLiNER.from_pretrained("anthonyyazdaniml/GLiNER-BioClinX")

text = "John Smith was prescribed metformin for type 2 diabetes."
labels = ["person name", "medication", "disease or condition"]

entities = model.predict_entities(text, labels, threshold=0.5)
for entity in entities:
    print(f"{entity['text']} -> {entity['label']}")

Expected output:

John Smith -> person name
metformin -> medication
type 2 diabetes -> disease or condition