| """BD3LM config for Hugging Face. |
| |
| """ |
|
|
| import transformers |
|
|
|
|
| class BD3LMConfig(transformers.PretrainedConfig): |
| """Hugging Face configuration class for BD3LM.""" |
| model_type = "bd3lm" |
|
|
| def __init__( |
| self, |
| block_size: int = 1, |
| vocab_size: int = 4107, |
| model_length: int = 1024, |
| cross_attn: bool = True, |
| adaln: bool = True, |
| attn_backend: str = 'flex', |
| causal: bool = False, |
| hidden_dim: int = 768, |
| cond_dim: int = 129, |
| n_blocks: int = 12, |
| n_heads: int = 12, |
| dropout: float = 0.1, |
| time_conditioning: bool = False, |
| var_min: bool = True, |
| sampling_eps_min: float = 1e-3, |
| sampling_eps_max: float = 0.999, |
| ** kwargs): |
| super().__init__(**kwargs) |
| self.block_size = block_size |
| self.cross_attn = cross_attn |
| self.adaln = adaln |
| self.attn_backend = attn_backend |
| self.causal = causal |
| self.vocab_size = vocab_size |
| self.model_length = model_length |
| self.hidden_dim = hidden_dim |
| self.cond_dim = cond_dim |
| self.n_blocks = n_blocks |
| self.n_heads = n_heads |
| self.dropout = dropout |
| self.time_conditioning = time_conditioning |
| self.var_min = var_min |
| self.sampling_eps_min = sampling_eps_min |
| self.sampling_eps_max = sampling_eps_max |
|
|
