File size: 3,099 Bytes
2cba4b1
f9714b1
2cba4b1
 
 
94de7c5
1c56cce
94de7c5
 
 
f9714b1
 
 
 
 
 
 
 
94de7c5
7538db6
1c56cce
9c36e3e
2cba4b1
 
 
 
 
 
 
 
 
 
 
 
 
94de7c5
 
 
f9714b1
 
 
 
94de7c5
84af19a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
94de7c5
 
 
f9714b1
 
 
94de7c5
 
 
 
 
 
 
 
 
f9714b1
94de7c5
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
import onnxruntime as ort
from transformers import AutoTokenizer
import numpy as np
import requests
import os

VERSION = "v0.1.1"
class IntentClassifier:
    def __init__(self):
        self.id2label = {0: 'information_intent',
                         1: 'yelp_intent',
                         2: 'navigation_intent',
                         3: 'travel_intent',
                         4: 'purchase_intent',
                         5: 'weather_intent',
                         6: 'translation_intent',
                         7: 'unknown'}
        self.label2id = {label: id for id, label in self.id2label.items()}

        self.tokenizer = AutoTokenizer.from_pretrained("Mozilla/mobilebert-uncased-finetuned-LoRA-intent-classifier")
        
        model_url = f"https://huggingface.co/Mozilla/mobilebert-uncased-finetuned-LoRA-intent-classifier/resolve/{VERSION}/onnx/model_quantized.onnx"
        model_dir_path = "models"
        model_path = f"{model_dir_path}/mobilebert-uncased-finetuned-LoRA-intent-classifier_model_quantized.onnx"
        if not os.path.exists(model_dir_path):
            os.makedirs(model_dir_path)
        if not os.path.exists(model_path):
            print("Downloading ONNX model...")
            response = requests.get(model_url)
            with open(model_path, "wb") as f:
                f.write(response.content)
            print("ONNX model downloaded.")

        # Load the ONNX model
        self.ort_session = ort.InferenceSession(model_path)

    def find_intent(self, sequence, verbose=False):
        inputs = self.tokenizer(sequence,
                                return_tensors="np",  # ONNX requires inputs in NumPy format
                                padding="max_length",  # Pad to max length
                                truncation=True,       # Truncate if the text is too long
                                max_length=64)

        # Convert inputs to NumPy arrays
        onnx_inputs = {k: v for k, v in inputs.items()}

        # Run the ONNX model
        logits = self.ort_session.run(None, onnx_inputs)[0]

        # Get the prediction
        prediction = np.argmax(logits, axis=1)[0]
        probabilities = np.exp(logits) / np.sum(np.exp(logits), axis=1, keepdims=True)
        rounded_probabilities = np.round(probabilities, decimals=3)

        pred_result = self.id2label[prediction]
        proba_result = dict(zip(self.label2id.keys(), rounded_probabilities[0].tolist()))
        
        if verbose:
            print(sequence + " -> " + pred_result)
            print(proba_result, "\n")
        
        return pred_result, proba_result

def main():
    text_list = [
        'floor repair cost',
        'pet store near me',
        'who is the us president',
        'italian food',
        'sandwiches for lunch',
        "cheese burger cost",
        "What is the weather today",
        "what is the capital of usa",
        "cruise trip to carribean",
    ]
    cls = IntentClassifier()
    for sequence in text_list:
        cls.find_intent(sequence, verbose=True)

if __name__ == '__main__':
    main()