This is a structured projection layer that projects the input to a higher dimension. It only allows integer up-projection factors, i.e. the output

Changes the weight decay value for a single parameter Use-cases: - ViT exclude CLS token parameters - Transformer learned positional

Efficiently drop paths (Stochastic Depth) per sample such that dropped samples are not processed. This is a subclass of nn.Sequential and can

LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False.

This is a structured projection layer that projects the input to a higher dimension. It only allows integer up-projection factors, i.e. the o

This is a structured projection layer that projects the input to a higher dimension. It only allows integer up-projection factors, i.e. the o

LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False.

LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False.

This class helps keeping parameters in a specialized internal structure to be optimal for computation speed, while having a proxied version t

wrapper for torchvision.transforms.Resize as it doesn't support passing a string as interpolation

Implements causal depthwise convolution of a time series tensor. Input: Tensor of shape (B,T,F), i.e. (batch, time, feature) Output: Ten

Implements causal depthwise convolution of a time series tensor. Input: Tensor of shape (B,T,F), i.e. (batch, time, feature) Output: Ten

hyperparams for specific properties in the dictionary and replace it such that the training duration is limited to a minimal configuration

limit training duration to a minimum by manipulating the configuration yaml

2D Image to Patch Embedding

remove large list/dicts for prettier storing of the resolved yaml

resolves nested lists like this: schedule: schedule: - kind: ... - kind: ... into this: schedule: - kind: