A residual block that can optionally change the number of channels. :param channels: the number of input channels. :param emb_channels:

Computes and stores the average and current value

A sequential module that passes timestep embeddings to the children that support it as an extra input.

Computes and stores the average and current value

Channel-wise self-attention block.

Residual block for StableGeneratorResnet Modified to remove ReflectionPad for deterministic behavior

An attention block that allows spatial positions to attend to each other. Originally ported from here, but adapted to the N-d case. http

forward: out = relu(x) backward: grad = gamma * relu(x)

A downsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determin

Tokens-to-Token encoding module

Vision Transformer with support for patch or hybrid CNN input stage

Maintains (exponential) moving average of a set of parameters.

An upsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determini

Multi-Head Attention

TNT Block

Standardize the channels of a batch of images by subtracting the dataset mean and dividing by the dataset standard deviation. In order to

Image to Patch Embedding

A module which performs QKV attention and splits in a different order.

Generates quantized convolutional layers. args: bit(int): bitwidth for the quantization, power(bool): (A)PoT or Uniform quantizat