| from transformers import PretrainedConfig, AutoConfig | |
| class BERTMultiGATAttentionConfig(PretrainedConfig): | |
| model_type = "deberta_semantic_similarity" | |
| def __init__( | |
| self, | |
| transformer_model='microsoft/deberta-v3-base', | |
| hidden_size=768, | |
| num_heads=8, | |
| dropout=0.07, | |
| gnn_input_dim=768, | |
| gnn_hidden_dim=768, | |
| **kwargs | |
| ): | |
| super().__init__(**kwargs) | |
| self.transformer_model = transformer_model | |
| self.hidden_size = hidden_size | |
| self.num_heads = num_heads | |
| self.dropout = dropout | |
| self.gnn_input_dim = gnn_input_dim | |
| self.gnn_hidden_dim = gnn_hidden_dim | |
| AutoConfig.register("deberta_semantic_similarity", BERTMultiGATAttentionConfig) | |