| from transformers import PretrainedConfig | |
| from typing import Optional | |
| class LlamaEdgeConfig(PretrainedConfig): | |
| model_type = "llama_edge" | |
| def __init__( | |
| self, | |
| dim: int = 4096, | |
| n_layers: int = 32, | |
| n_heads: int = 32, | |
| n_kv_heads: int = 8, | |
| vocab_size: int = 9942, | |
| multiple_of: int = 256, | |
| ffn_dim_multiplier: Optional[float] = 1.3, | |
| norm_eps: float = 1e-5, | |
| rope_theta: float = 500000.0, | |
| max_seq_len: int = 8192, | |
| intermediate_size: int = 14336, | |
| **kwargs, | |
| ): | |
| self.dim = dim | |
| self.n_layers = n_layers | |
| self.n_heads = n_heads | |
| self.n_kv_heads = n_kv_heads | |
| self.vocab_size = vocab_size | |
| self.multiple_of = multiple_of | |
| self.ffn_dim_multiplier = ffn_dim_multiplier | |
| self.norm_eps = norm_eps | |
| self.rope_theta = rope_theta | |
| self.max_seq_len = max_seq_len | |
| self.intermediate_size = intermediate_size | |
| super().__init__(**kwargs) | |