cerebras.modelzoo.data.vision.diffusion.DiffusionLatentImageNet1KProcessor.DiffusionLatentImageNet1KProcessor#
- class cerebras.modelzoo.data.vision.diffusion.DiffusionLatentImageNet1KProcessor.DiffusionLatentImageNet1KProcessor(params)[source]#
Bases:
cerebras.modelzoo.data.vision.diffusion.DiffusionBaseProcessor.DiffusionBaseProcessor
Methods
check_split_valid
Dataloader returns a dict with keys:
create_dataset
custom_collate_fn
process_transform
- create_dataloader(dataset, is_training=False)#
- Dataloader returns a dict with keys:
“input”: Tensor of shape (batch_size, latent_channels, latent_height, latent_width) “label”: Tensor of shape (batch_size, ) with dropout applied with label_dropout_rate “diffusion_noise”: Tensor of shape (batch_size, latent_channels, latent_height, latent_width)
represents diffusion noise to be applied
- “timestep”: Tensor of shape (batch_size, ) that
indicates the timesteps for each diffusion sample