| from transformers import AutoTokenizer, AutoModel | |
| import torch | |
| import os | |
| os.environ['HF_HOME'] = 'model/' | |
| os.environ['TRANSFORMERS_CACHE'] = 'model/transformers_cache' | |
| os.environ['HUGGINGFACE_HUB_CACHE'] = 'model/transformers_cache' | |
| class CrossEncoderBert(torch.nn.Module): | |
| def __init__(self): | |
| super().__init__() | |
| self.model = AutoModel.from_pretrained('distilbert-base-uncased') | |
| self.tokenizer = AutoTokenizer.from_pretrained('distilbert-base-uncased') | |
| self.tokenizer.add_tokens(["[Cont_token]"], special_tokens=True) | |
| self.linear = torch.nn.Linear(self.model.config.hidden_size, 1) | |
| def forward(self, input_ids, attention_mask): | |
| outputs = self.model(input_ids=input_ids, attention_mask=attention_mask) | |
| pooled_output = outputs.last_hidden_state[:, 0] | |
| return self.linear(pooled_output) | |