| from torch import nn | |
| from transformers import BertModel,BertConfig | |
| from transformers.modeling_outputs import TokenClassifierOutput | |
| class BertClassifier(nn.Module): | |
| def __init__(self, num_labels=2, dropout=0.1,bert_model=None): | |
| super().__init__() | |
| if bert_model: | |
| self.bert = BertModel.from_pretrained(bert_model) | |
| else: | |
| config = BertConfig(vocab_size=34688, max_position_embeddings=512) | |
| self.bert = BertModel(config=config) | |
| self.num_labels = num_labels | |
| self.classifier = nn.Sequential( | |
| nn.Linear(self.bert.config.hidden_size, self.bert.config.hidden_size), | |
| nn.ReLU(), | |
| nn.Dropout(dropout), | |
| nn.Linear(self.bert.config.hidden_size, num_labels)) | |
| def forward(self, input_ids=None, attention_mask=None,labels=None): | |
| output = self.bert(input_ids, attention_mask=attention_mask) | |
| logits = self.classifier(output.pooler_output) | |
| loss = None | |
| if labels: | |
| loss_fct = nn.CrossEntropyLoss() | |
| loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) | |
| return TokenClassifierOutput(loss=loss, logits=logits, hidden_states=output.hidden_states,attentions=output.attentions) | |