这是一个用于分析句子情感的模型。

该模型基于KcELECTRA，使用约22万个情感句子进行了训练。

将原有的FP16模型转换为ONNX FP16格式，使CPU环境下的推理速度提升了约2倍。

情感共分为6个类别：喜悦、困惑、愤怒、不安、受伤、悲伤。

# 转换为ONNX时softmax被移除，因此使用numpy重新实现
def softmax(x):
    x = np.array(x)
    e_x = np.exp(x - np.max(x))
    return e_x / e_x.sum(axis=-1, keepdims=True)

def predict(text: str):
    text = clean(text)

    inputs = tokenizer(text, return_tensors=\"np\")

    ort_inputs = {
        \"input_ids\": inputs[\"input_ids\"],
        \"attention_mask\": inputs[\"attention_mask\"]
    }

    logits = session.run([\"logits\"], ort_inputs)[0]  # (1, 6)
    probs = softmax(logits)[0]  # shape: (6,)

    result = {id2label[i]: float(probs[i]) for i in range(6)}
    return result

# pip install optimum[onnxruntime]

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer

model_id = \"noridorimari/onnx_emotion_classifier\"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = ORTModelForSequenceClassification.from_pretrained(model_id)

inputs = tokenizer(\"今天心情不太好。\", return_tensors=\"pt\")
outputs = model(**inputs)
print(outputs.logits)

@misc{lee2021kcelectra,
  author = {Junbum Lee},
  title = {KcELECTRA: Korean comments ELECTRA},
  year = {2021},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/Beomi/KcELECTRA}}
}