SDDOIA DPL loss

Addminst DPL loss class

Diffusion model with a Transformer backbone.

Kandinksy DPL loss

Subclass torch's LayerNorm to handle fp16.

Returns a compatible Torch Dataset object customized for the BDD dataset

Returns a compatible Torch Dataset object customized for the BDD dataset

nMNIST dataset.

This model adds conditioning to y when generating w, essentially interleaving their computation.

Custom Dataset class for performing operations on MNIST images. This class creates a dataset where each sample consists of multiple images (

XOR DPL loss

Dataset for concept image loading

This model adds conditioning to y when generating w, essentially interleaving their computation.

XOR DPL loss

nMNIST dataset.

nMNIST dataset.

def __init__: define parameters (e.g., # of layers) MLP-based conceptizer for concept basis learning. Inputs: din (in

This model mimics the conceptizer from BEARS https://github.com/samuelebortolotti/bears/blob/master/BDD_OIA/conceptizers_BDD.py def __init__:

Fully neural MODEL FOR BOIA

CBM MODEL FOR BOIA

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

Taken from https://github.com/kuleshov-group/mdlm/blob/master/models/dit.py#L292

Fully neural MODEL FOR Kandinsky

https://github.com/kuleshov-group/mdlm/blob/master/models/dit.py#L126

Fully neural MODEL FOR MNIST

Fully neural MODEL FOR MNMATH

CBM MODEL FOR MNIST

A ResNet class that is similar to torchvision's but contains the following changes: - There are now 3 "stem" convolutions as opposed to 1, wi