The implementation of SCNet: Sparse Compression Network for Music Source Separation. Paper: https://arxiv.org/abs/2401.13276.pdf Args: -

AdaGrad Meets Muon: Adaptive Stepsizes for Orthogonal Updates. Args: params (Parameters): The parameters to be optimized by Muon.

Momentum Orthogonalized by Newton-schulz. Muon internally runs standard SGD-momentum, and then performs an orthogonalization post-processing step

Implements a simplified Sparse Down-sample block in an encoder architecture. Args: - channels (int): Number input channels. - expand

Adaptive Muon optimizer. Muon internally runs standard SGD-momentum, and then performs an orthogonalization post-processing step, in which ea

Совместимо с твоим train: forward(x: (B, C, T)) -> y_hat: (B, S, C, T) где S = число источников (sources). Внутри: STFT -> NeuralMo

Convolution Module in SD block. Args: channels (int): input/output channels. depth (int): number of layers in the residual b

Convolution Module in SD block. Args: channels (int): input/output channels. depth (int): number of layers in the residual b

Convolution Module in SD block. Args: channels (int): input/output channels. depth (int): number of layers in the residual b

ConvolutionModule class implements a module with a sequence of convolutional layers similar to Conformer. Args: - input_dim (int): Dimen

*Dual-path recurrent neural networks (DPRNN)* :cite:`luo2020dual`. Args: feat_dim (int, optional): The feature dimension after Encoder mo