models.py

import torch
from transformers import BertTokenizer
from torch import nn
from transformers import BertModel


#CustomMBERTModel= torch.load("/data2/Akash_for_interface/model_mbert_1416.pt")
tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
sentences= "you are good  person."
max_len=len(sentences)
encoding = tokenizer.encode_plus(sentences,add_special_tokens=True,max_length=max_len,padding='max_length',truncation=True,return_tensors='pt')
input_ids=encoding['input_ids'].flatten(),
attention_mask= encoding['attention_mask'].flatten()
#print(input_ids[0])

labels=["Non Hateful","Hateful"]



device = 'cpu'
class CustomMBERTModel(nn.Module):
    def __init__(self, num_labels):
        super(CustomMBERTModel, self).__init__()
        self.bert = BertModel.from_pretrained('bert-base-multilingual-cased')
        
        # Freeze all layers except the top 2
        for param in self.bert.parameters():
            param.requires_grad = False

        # Unfreeze the parameters of the top 2 layers
        for param in self.bert.encoder.layer[-2:].parameters():
            param.requires_grad = True

        # Adding Linear layer
        self.linear = nn.Linear(self.bert.config.hidden_size, num_labels)

    def forward(self, input_ids, attention_mask):
        outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
        hidden_states = outputs.last_hidden_state

        pooled_output= hidden_states[:,0,:]
        pooled_output=torch.squeeze(pooled_output,dim=1)

        #print('p-shape:', pooled_output.shape)

        logits = self.linear(pooled_output)

        #print('l-shape:', logits.shape)
        return logits