thomasbeste/modernbert-da-ner-base-onnx-int8

ModernBERT Danish NER (Base) — ONNX INT8

ONNX INT8 dynamically quantized version of thomasbeste/modernbert-da-ner-base.

Quantized with AVX-512 VNNI configuration for fast CPU inference.

Benchmark: DaNE Test Set

Entity	Precision	Recall	F1	Support
PER	0.8962	0.9061	0.9011	181
ORG	0.6929	0.6299	0.6599	154
LOC	0.7500	0.8969	0.8169	97
MISC	0.4878	0.6316	0.5505	95
micro avg	0.7260	0.7742	0.7493

Usage

from optimum.onnxruntime import ORTModelForTokenClassification
from transformers import AutoTokenizer, pipeline

model = ORTModelForTokenClassification.from_pretrained("thomasbeste/modernbert-da-ner-base-onnx-int8")
tokenizer = AutoTokenizer.from_pretrained("thomasbeste/modernbert-da-ner-base-onnx-int8")
ner = pipeline("ner", model=model, tokenizer=tokenizer, aggregation_strategy="simple")

results = ner("Jens Peter Hansen bor i København og arbejder hos Novo Nordisk.")
for entity in results:
    print(f"{entity['word']}: {entity['entity_group']} ({entity['score']:.3f})")

Training Details

See the PyTorch model card: thomasbeste/modernbert-da-ner-base