from transformers import RobertaTokenizer, RobertaModel
from transformers import BertModel, BertTokenizer
from torch import nn
    
#================
# BERT MODEL
#================
BERT_MODEL_PATH = "./BERT_MODEL.pth"
bert_tokenizer=BertTokenizer.from_pretrained('bert-base-uncased')
class BertForMultiLabel(nn.Module):
    def __init__(self, pretrained_model='bert-base-uncased', num_labels=5):
        super().__init__()
        self.bert = BertModel.from_pretrained(pretrained_model)
        self.dropout = nn.Dropout(0.3)
        self.classifier = nn.Linear(self.bert.config.hidden_size, num_labels)

    def forward(self, input_ids, attention_mask):
        pooled_output = self.bert(input_ids=input_ids, attention_mask=attention_mask).pooler_output
        return self.classifier(self.dropout(pooled_output))