models/BertTunning.py

import pandas as pd
import numpy as np
import torch
from torch import nn
import torch.nn.functional as F


from logreg_model import bert_for_logreg, tokenizer_bert
from preprocess_bert import preprocess_bert

MAX_LEN = 100

class BertTunnig(nn.Module):

    def __init__(self, bert_model):
        super().__init__()

        self.bert = bert_model
        for weights in self.bert.parameters():
            weights.requires_grad = False

        self.fc1 = nn.Linear(768, 256)
        self.drop1 = nn.Dropout(p=0.5)
        self.fc2 = nn.Linear(256, 32)
        self.fc_out = nn.Linear(32, 1)

    def forward(self, x, attention_mask):

        output = self.bert(x, attention_mask=attention_mask)[0][:, 0, :]
        output = self.fc1(output)

        output_drop = self.drop1(output)
        output = self.fc2(output_drop)

        output = self.fc_out(output)

        return torch.sigmoid(output)


model_tunning = BertTunnig(bert_model=bert_for_logreg)

model_tunning.load_state_dict(torch.load('best_weights_berttinnug(2).pt'))

def predict_2(text):

    preprocessed_text, attention_mask = preprocess_bert(text, MAX_LEN=MAX_LEN)
    preprocessed_text, attention_mask = torch.tensor(preprocessed_text).unsqueeze(0), torch.tensor([attention_mask])

    with torch.inference_mode():

        predict = model_tunning(preprocessed_text, attention_mask=attention_mask).item()

    return round(predict)