| from transformers import PretrainedConfig | |
| class SpecT1Config(PretrainedConfig): | |
| model_type = "spect1" | |
| def __init__( | |
| self, | |
| hidden_size=4096, | |
| intermediate_size=11008, | |
| num_attention_heads=32, | |
| num_hidden_layers=36, | |
| max_position_embeddings=32768, | |
| vocab_size=151680, | |
| attention_dropout=0.0, | |
| hidden_act="silu", | |
| max_window_layers=36, | |
| rope_theta=640000, | |
| sliding_window=32768, | |
| attention_bias=True, | |
| num_nextn_predict_layers=1, | |
| initializer_range=0.02, | |
| bos_token_id=151643, | |
| eos_token_id=151645, | |
| tie_word_embeddings=False, | |
| **kwargs | |
| ): | |
| super().__init__(**kwargs) | |
| self.hidden_size = hidden_size | |
| self.intermediate_size = intermediate_size | |
| self.num_attention_heads = num_attention_heads | |
| self.num_hidden_layers = num_hidden_layers | |
| self.max_position_embeddings = max_position_embeddings | |
| self.vocab_size = vocab_size | |
| self.attention_dropout = attention_dropout | |
| self.hidden_act = hidden_act | |
| self.max_window_layers = max_window_layers | |
| self.rope_theta = rope_theta | |
| self.sliding_window = sliding_window | |
| self.attention_bias = attention_bias | |
| self.num_nextn_predict_layers = num_nextn_predict_layers | |
| self.initializer_range = initializer_range | |
| self.bos_token_id = bos_token_id | |
| self.eos_token_id = eos_token_id | |
| self.tie_word_embeddings = tie_word_embeddings | |