Task: Token classification (NER) with BIO tagging
Base model: microsoft/mdeberta-v3-base
Entity types (18): PERSON, EMAIL, PHONE, ADDRESS, LOCATION, ID_NUMBER, CREDIT_CARD, DATE_OF_BIRTH, ORGANIZATION, IP_ADDRESS, URL, USERNAME, PASSWORD, MEDICAL_ID, BANK_ACCOUNT, SECRET, DEVICE_ID, DEMOGRAPHIC
Languages: 11 (en, vi, hi, th, zh, ja, ru, ar, sv, es, it)
Export formats: ONNX FP32 + FP16 + INT8 (x86-safe dynamic)

Performance

Metric	Score
loss	0.1339
precision	0.8646
recall	0.8881
f1	0.8762

Usage (ONNX)

python

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

model = ORTModelForTokenClassification.from_pretrained(
    "HikmaAI/hikmaai-mdeberta-v3-base-pii-multilingual",
    subfolder="onnx/fp16",
)
tokenizer = AutoTokenizer.from_pretrained(
    "HikmaAI/hikmaai-mdeberta-v3-base-pii-multilingual",
    subfolder="tokenizer",
)

ner = pipeline(
    "token-classification",
    model=model,
    tokenizer=tokenizer,
    aggregation_strategy="simple",
)
ner("My name is Jane Doe and my email is jane@example.com")

Training

Epochs: 4
Learning rate: 2e-05
Batch size: 16
Dataset: multilingual (11 languages), Nemotron-PII + ai4privacy 300k + gretelai + WikiANN

License

Apache-2.0

Citation

bibtex

@misc{hikmaai-pii-2026,
  title={hikmaai-mdeberta-v3-base-pii-multilingual},
  author={HikmaAI},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/HikmaAI/hikmaai-mdeberta-v3-base-pii-multilingual}
}