Views
No views yet
1import numpy as np
2import onnxruntime as ort
3from transformers import AutoTokenizer
4
5def softmax(x):
6 exp_x = np.exp(x - np.max(x, axis=1, keepdims=True))
7 return exp_x / np.sum(exp_x, axis=1, keepdims=True)
8
9session = ort.InferenceSession(
10 "chinese_best_model_q8.onnx", providers=["CPUExecutionProvider"]
11)
12
13tokenizer = AutoTokenizer.from_pretrained(
14 "./tokenizer",
15 local_files_only=True,
16 truncation_side="left"
17)
18
19text = "这是一句没有标点的文本"
20inputs = tokenizer(
21 text,
22 truncation=True,
23 padding='max_length',
24 add_special_tokens=False,
25 return_tensors="np",
26 max_length=128,
27 )
28# Run inference
29outputs = session.run(None,
30 {
31 "input_ids": inputs["input_ids"].astype("int64"),
32 "attention_mask": inputs["attention_mask"].astype("int64")
33 })
34eou_probability = softmax(outputs[0]).flatten()[-1]
35print(eou_probability, eou_probability>0.5)