Diffusion model with a Transformer backbone.

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

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

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

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

Vision Transformer

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

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

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

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

Vision Transformer

Masked Autoencoder with VisionTransformer backbone

Image to Patch Embedding

Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard transform

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

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

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

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

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

Track a series of values and provide access to smoothed values over a window or the global series average.

see https://github.com/MishaLaskin/vqvae/blob/d761a999e2267766400dc646d82d3ac3657771d4/models/quantizer.py __________________________________

Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py

Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py

Uses a pretrained BERT tokenizer by huggingface. Vocab size: 30522 (?)

Construct the embeddings from word, position and token_type embeddings.

A DiT block with adaptive layer norm zero (adaLN-Zero) conditioning.

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

The half UNet model with attention and timestep embedding. For usage, see UNet.

The final layer of DiT.

Utilities for training and sampling diffusion models. Ported directly from here, and then adapted over time to further experimentation.

Utilities for training and sampling diffusion models. Original ported from this codebase: https://github.com/hojonathanho/diffusion/blob/

Embeds class labels into vector representations. Also handles label dropout for classifier-free guidance.

NLL loss with label smoothing.

to match AttnBlock usage

Image to Patch Embedding

A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping

A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping

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

A diffusion process which can skip steps in a base diffusion process. :param use_timesteps: a collection (sequence or set) of timesteps from

A diffusion process which can skip steps in a base diffusion process. :param use_timesteps: a collection (sequence or set) of timesteps from

A UNetModel that performs super-resolution. Expects an extra kwarg `low_res` to condition on a low-resolution image.

Dumps key/value pairs into TensorBoard's numeric format.

Embeds scalar timesteps into vector representations.

The full UNet model with attention and timestep embedding. :param in_channels: channels in the input Tensor. :param model_channels: base