main.py

from flask import Flask, request, jsonify
import torch
from transformers import RobertaTokenizer
import os
from transformers import RobertaForSequenceClassification
import torch.serialization
import torch
from transformers import RobertaTokenizer, RobertaForSequenceClassification, Trainer, TrainingArguments
from torch.utils.data import Dataset
import pandas as pd
from sklearn.model_selection import train_test_split
import numpy as np
# Initialize Flask app
app = Flask(__name__)

# Load the trained model and tokenizer
tokenizer = RobertaTokenizer.from_pretrained("microsoft/codebert-base")
torch.serialization.add_safe_globals([RobertaForSequenceClassification])

model = torch.load("model.pth", map_location=torch.device('cpu'), weights_only=False)  # Load the trained model

# Ensure the model is in evaluation mode
model.eval()


@app.route("/")
def home():
    return request.url


# @app.route("/predict", methods=["POST"])
@app.route("/predict")
def predict():
    print("Received code:", request.get_json()["code"])
    code = request.get_json()["code"]
    # Load saved weights and config
    checkpoint = torch.load("codebert_vulnerability_scorer.pth")
    config = RobertaConfig.from_dict(checkpoint['config'])

    # Rebuild the model with correct architecture
    model = RobertaForSequenceClassification(config)
    model.load_state_dict(checkpoint['model_state_dict'])
    model.eval()

    # Load tokenizer
    tokenizer = RobertaTokenizer.from_pretrained('./tokenizer')

    # Prepare input
    inputs = tokenizer(
        code,
        truncation=True,
        padding='max_length',
        max_length=512,
        return_tensors='pt'
    )

    # Make prediction
    with torch.no_grad():
        outputs = model(**inputs)

    score = torch.sigmoid(outputs.logits).item()
    return score
    


    
# Run the Flask app
if __name__ == "__main__":
     app.run(host="0.0.0.0", port=7860)