This is an OpenMed token-classification checkpoint intended for Japanese (ja) personally identifiable information (PII) and protected health information (PHI) span detection.

Model details

Language scope: Japanese (ja)
Task: token classification / named entity recognition
Base model: lightonai/modernbert-embed-large
Library: Transformers

Usage

python

from transformers import pipeline

model_id = "OpenMed/OpenMed-PII-Japanese-NomicMed-Large-395M-v1"
detector = pipeline(
    "token-classification",
    model=model_id,
    aggregation_strategy="simple",
)

text = '架空の患者、山田太郎さんのメールアドレスは taro.yamada@example.com、電話番号は 090-0000-0000 です。'
print(detector(text))

The checkpoint's configured id2label mapping is authoritative for the available entity labels. Preserve returned character offsets when applying redaction or replacement.

Evaluation status

No verified Japanese evaluation artifact was available during this metadata repair, so this card intentionally reports no language-specific scores. Evaluate direct-identifier recall, false negatives, span boundaries, and domain shift on representative data before deployment.

Limitations and safety

This model can miss identifiers or over-redact clinically useful context. It is not an anonymization guarantee, a compliance determination, or a medical device. Use defense in depth and human review for high-sensitivity workflows. Do not include real patient information in public examples, logs, or issue reports.