| from transformers import PretrainedConfig |
|
|
| class ConfigHybridBiGRUModel(PretrainedConfig): |
| model_type = "bert-bigru" |
|
|
| def __init__(self, |
| bert_model_name="bert-base-uncased", |
| tokenizer_name="bert-base-uncased", |
| hidden_dim=128, |
| num_classes=2, |
| gru_layers=1, |
| bidirectional=True, |
| dropout=0.3, |
| concat_layers=0, |
| pooling="last", |
| freeze_bert=False, |
| freeze_n_layers=0, |
| freeze_from_start=False, |
| **kwargs): |
| super().__init__(**kwargs) |
| self.bert_model_name = bert_model_name |
| self.tokenizer_name = tokenizer_name |
| self.hidden_dim = hidden_dim |
| self.num_classes = num_classes |
| self.gru_layers = gru_layers |
| self.bidirectional = bidirectional |
| self.dropout = dropout |
| self.concat_layers = concat_layers |
| self.pooling = pooling |
| self.freeze_bert = freeze_bert |
| self.freeze_n_layers = freeze_n_layers |
| self.freeze_from_start = freeze_from_start |
