from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

model_id = "jordigonzm/mdeberta-v3-base-multilingual-ner" # 轻量级模型示例
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForTokenClassification.from_pretrained(model_id)

# 创建 NER 管道
# 'aggregation_strategy' 将子词组合成完整单词
nlp_ner = pipeline("ner", model=model, tokenizer=tokenizer, aggregation_strategy="simple")

text = "La sede de Microsoft se encuentra en Redmond, Washington."
results = nlp_ner(text)

for entity in results:
    print(f"实体: {entity['word']} | 类型: {entity['entity_group']} | 得分: {entity['score']:.4f}")

导出为 ONNX

optimum-cli export onnx --model jordigonzm/mdeberta-v3-base-multilingual-ner --task token-classification --optimize O2 hf_mdeberta_ner
optimum-cli onnxruntime quantize --onnx_model hf_mdeberta_ner -o onnx_mdeberta_ner --avx512_vnni --per_channel