| from transformers import PretrainedConfig | |
| from typing import Sequence | |
| class DMAE1dConfig(PretrainedConfig): | |
| model_type = "archinetai/dmae1d-ATC64-v1" | |
| def __init__( | |
| self, | |
| in_channels: int = 2, | |
| channels: int = 512, | |
| multipliers: Sequence[int] = [3, 2, 1, 1, 1, 1, 1, 1], | |
| factors: Sequence[int] = [1, 2, 2, 2, 2, 2, 2], | |
| num_blocks: Sequence[int] = [1, 1, 1, 2, 2, 2, 2], | |
| attentions: Sequence[int] = [0, 0, 0, 0, 0, 0, 0], | |
| encoder_inject_depth: int = 3, | |
| encoder_channels: int = 32, | |
| encoder_factors: Sequence[int] = [1, 1, 2, 2, 1, 1], | |
| encoder_multipliers: Sequence[int] = [32, 16, 8, 8, 4, 2, 1], | |
| encoder_num_blocks: Sequence[int] = [4, 4, 4, 4, 4, 4], | |
| bottleneck: str = 'tanh', | |
| stft_use_complex: bool = True, | |
| stft_num_fft: int = 1023, | |
| stft_hop_length: int = 256, | |
| **kwargs | |
| ): | |
| self.in_channels = in_channels | |
| self.channels = channels | |
| self.multipliers = multipliers | |
| self.factors = factors | |
| self.num_blocks = num_blocks | |
| self.attentions = attentions | |
| self.encoder_inject_depth = encoder_inject_depth | |
| self.encoder_channels = encoder_channels | |
| self.encoder_factors = encoder_factors | |
| self.encoder_multipliers = encoder_multipliers | |
| self.encoder_num_blocks = encoder_num_blocks | |
| self.bottleneck = bottleneck | |
| self.stft_use_complex = stft_use_complex | |
| self.stft_num_fft = stft_num_fft | |
| self.stft_hop_length = stft_hop_length | |
| super().__init__(**kwargs) | |