| from torch import nn |
| from easydict import EasyDict as MyEasyDict |
| from transformers import BertModel, PreTrainedModel, BertConfig, PretrainedConfig |
|
|
|
|
| class BertConfig(PretrainedConfig): |
| model_type = "bert" |
|
|
| def __init__( |
| self, |
| model_config=None, |
| **kwargs): |
| super().__init__(**kwargs) |
| self.model_config = MyEasyDict(model_config) |
|
|
|
|
| class BERTClassifier(PreTrainedModel): |
| config_class = BertConfig |
|
|
| def __init__(self, config): |
| super().__init__(config) |
| self.bert = BertModel(config) |
| self.dropout = nn.Dropout(0.1) |
| self.fc = nn.Linear(self.bert.config.hidden_size, 16) |
|
|
| def forward(self, input_ids, attention_mask): |
| outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask) |
| pooled_output = outputs.pooler_output |
| x = self.dropout(pooled_output) |
| logits = self.fc(x) |
| return logits |
| |
| def print_test(self, x): |
| return "lmao" |
|
|
|
|
| if __name__ == "__main__": |
|
|
| from transformers import BertConfig, BertModel, BertForMaskedLM, AutoConfig |
|
|
| |
| config = AutoConfig.from_pretrained('google-bert/bert-base-uncased', trust_remote_code=True) |
| model = BERTClassifier(config) |