| |
| from transformers import PretrainedConfig |
|
|
|
|
| class RoseX15Config(PretrainedConfig): |
| |
| model_type = "rose_x1_5" |
|
|
| def __init__( |
| self, |
| vocab_size=32768, |
| hidden_size=576, |
| num_hidden_layers=22, |
| num_attention_heads=9, |
| num_key_value_heads=3, |
| head_dim=64, |
| intermediate_size=1532, |
| max_position_embeddings=4096, |
| hidden_act="silu", |
| rms_norm_eps=1e-6, |
| attention_bias=False, |
| mlp_bias=False, |
| attention_dropout=0.0, |
| rope_theta=100000.0, |
| rope_scaling=None, |
| tie_word_embeddings=True, |
| |
| use_qk_norm=True, |
| |
| use_xsa=True, |
| |
| refresh_gate_enabled=True, |
| refresh_gate_inject_layers=None, |
| refresh_gate_kernel_size=9, |
| |
| mup_enabled=False, |
| mup_base_hidden_size=None, |
| mup_width_mult=None, |
| **kwargs, |
| ): |
| self.vocab_size = vocab_size |
| self.hidden_size = hidden_size |
| self.num_hidden_layers = num_hidden_layers |
| self.num_attention_heads = num_attention_heads |
| self.num_key_value_heads = num_key_value_heads |
| self.head_dim = head_dim |
| self.intermediate_size = intermediate_size |
| self.max_position_embeddings = max_position_embeddings |
| self.hidden_act = hidden_act |
| self.rms_norm_eps = rms_norm_eps |
| self.attention_bias = attention_bias |
| self.mlp_bias = mlp_bias |
| self.attention_dropout = attention_dropout |
| self.rope_theta = rope_theta |
| self.rope_scaling = rope_scaling |
| |
| self.use_qk_norm = use_qk_norm |
| |
| self.use_xsa = use_xsa |
| |
| self.refresh_gate_enabled = refresh_gate_enabled |
| self.refresh_gate_inject_layers = ( |
| list(refresh_gate_inject_layers) |
| if refresh_gate_inject_layers is not None |
| else [4, 9] |
| ) |
| self.refresh_gate_kernel_size = refresh_gate_kernel_size |
| |
| self.mup_enabled = mup_enabled |
| self.mup_base_hidden_size = mup_base_hidden_size |
| self.mup_width_mult = mup_width_mult |
| super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs) |
