Federated Diffusion Model with Non-IID Data

Setup

pip install labml-nn
pip install avalanche-lib

Training on federated setting

python federated.py

Training on centralized setting

python centralized.py

Set configurations.

You can override the defaults by passing the values in the function main().

For example:

experiment.configs(configs, {
    'dataset': 'MNIST',  # 'MNIST','CelebA'
    'image_channels': 1,  # 1,3
    'epochs': 5,  # 5,100
})

Sampling

This file can generate samples and interpolations from a trained model (You must run this evaluating after training a model).

python evaluate.py

Specify an uuid from trained logs in the function main().

run_uuid = "858546e54c6c11ed9b3764d69af93a2f"