utilities.py

import numpy as np
import pickle
import torch
import torch.nn as nn
from transformer import Transformer_Model
from torchtext.data import get_tokenizer
from torchtext.vocab import GloVe
    
model = Transformer_Model()
model = torch.jit.load('models/text_classification-0.1.0.pth')
model.eval()

global_vectors = GloVe(name='6B', dim=50)
tokenizer = get_tokenizer("basic_english")
padding = global_vectors.get_vecs_by_tokens(["."], lower_case_backup=True)

def preprocess(text):

    e = global_vectors.get_vecs_by_tokens(tokenizer(text), lower_case_backup=True)
    embeds = np.empty((100, 50))

    if(len(e) <= 100):

        embeds[0:len(e),:] = e
        embeds[len(e):100, :] = padding

    else:
        embeds = e[0:100,:]

    embeds = torch.tensor(embeds).float()
    embeds = torch.reshape(embeds, (1, 100, 50)).float()
    
    return embeds

def predict(embeddings):

    pred = model(embeddings)
    pred = torch.mean(pred)

    if(pred >= 0.5):
        return "Human written text"
    else:
        return "ChatGPT generated text"
    

def predict_pipeline(text):
    embeds = preprocess(text)
    prediction = predict(embeds)

    return prediction