Spaces:
Running
Running
Create main.py
Browse files
main.py
ADDED
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from flask import Flask, request, jsonify
|
2 |
+
import torch
|
3 |
+
from transformers import RobertaTokenizer
|
4 |
+
import os
|
5 |
+
from transformers import RobertaForSequenceClassification
|
6 |
+
import torch.serialization
|
7 |
+
import torch
|
8 |
+
from transformers import RobertaTokenizer, RobertaForSequenceClassification, Trainer, TrainingArguments
|
9 |
+
from torch.utils.data import Dataset
|
10 |
+
import pandas as pd
|
11 |
+
from sklearn.model_selection import train_test_split
|
12 |
+
import numpy as np
|
13 |
+
# Initialize Flask app
|
14 |
+
app = Flask(__name__)
|
15 |
+
|
16 |
+
# Load the trained model and tokenizer
|
17 |
+
tokenizer = RobertaTokenizer.from_pretrained("microsoft/codebert-base")
|
18 |
+
torch.serialization.add_safe_globals([RobertaForSequenceClassification])
|
19 |
+
|
20 |
+
model = torch.load("model.pth", map_location=torch.device('cpu'), weights_only=False) # Load the trained model
|
21 |
+
|
22 |
+
# Ensure the model is in evaluation mode
|
23 |
+
model.eval()
|
24 |
+
|
25 |
+
|
26 |
+
@app.route("/")
|
27 |
+
def home():
|
28 |
+
return request.url
|
29 |
+
|
30 |
+
|
31 |
+
# @app.route("/predict", methods=["POST"])
|
32 |
+
@app.route("/predict")
|
33 |
+
def predict():
|
34 |
+
print("Received code:", request.get_json()["code"])
|
35 |
+
code = request.get_json()["code"]
|
36 |
+
# Load saved weights and config
|
37 |
+
checkpoint = torch.load("codebert_vulnerability_scorer.pth")
|
38 |
+
config = RobertaConfig.from_dict(checkpoint['config'])
|
39 |
+
|
40 |
+
# Rebuild the model with correct architecture
|
41 |
+
model = RobertaForSequenceClassification(config)
|
42 |
+
model.load_state_dict(checkpoint['model_state_dict'])
|
43 |
+
model.eval()
|
44 |
+
|
45 |
+
# Load tokenizer
|
46 |
+
tokenizer = RobertaTokenizer.from_pretrained('./tokenizer')
|
47 |
+
|
48 |
+
# Prepare input
|
49 |
+
inputs = tokenizer(
|
50 |
+
code,
|
51 |
+
truncation=True,
|
52 |
+
padding='max_length',
|
53 |
+
max_length=512,
|
54 |
+
return_tensors='pt'
|
55 |
+
)
|
56 |
+
|
57 |
+
# Make prediction
|
58 |
+
with torch.no_grad():
|
59 |
+
outputs = model(**inputs)
|
60 |
+
|
61 |
+
score = torch.sigmoid(outputs.logits).item()
|
62 |
+
return score
|
63 |
+
|
64 |
+
|
65 |
+
|
66 |
+
|
67 |
+
# Run the Flask app
|
68 |
+
if __name__ == "__main__":
|
69 |
+
app.run(host="0.0.0.0", port=7860)
|