| |
| from transformers import PretrainedConfig |
|
|
|
|
| class SharedReconstructorConfig(PretrainedConfig): |
| model_type = "gpt2-shared-reconstructor" |
|
|
| def __init__(self, teacher_model_name="gpt2", hidden_size=768, |
| num_teacher_layers=12, max_position_embeddings=1024, |
| vocab_size=50257, num_attention_heads=12, ffn_size=3072, |
| layer_embedding_size=128, dropout=0.0, layer_norm_eps=1e-5, |
| **kwargs): |
| super().__init__(**kwargs) |
| self.teacher_model_name = teacher_model_name |
| self.hidden_size = hidden_size |
| self.num_teacher_layers = num_teacher_layers |
| self.max_position_embeddings = max_position_embeddings |
| self.vocab_size = vocab_size |
| self.num_attention_heads = num_attention_heads |
| self.ffn_size = ffn_size |
| self.layer_embedding_size = layer_embedding_size |
| self.dropout = dropout |
| self.layer_norm_eps = layer_norm_eps |
| self.architectures = ["HFSharedReconstructor"] |
| self.auto_map = { |
| "AutoConfig": "configuration_shared_reconstructor.SharedReconstructorConfig", |
| "AutoModel": "modeling_shared_reconstructor.HFSharedReconstructor", |
| } |
|
|
