Convex Diffusion: Interpolation for Dynamics Forecasting with Diffusion Models

Environment

Please follow the instructions in the environment folder to set up the correct conda environment.

Training a model

From the repository root, please run the following in the command line:

python run.py trainer.gpus=0 model=unet_convnext logger=none callbacks=default

This will train a UNet ConvNext on the CPU using some default callbacks and hyperparameters, but no logging. To change the used data directory you can override the flag datamodule.data_dir=YOUR-DATA-DIR.

*For more configuration & training options, please see the src/README.

Compute

• Please follow the instructions in the nautilus folder to set up the repository on the Nautilus cluster.
• Please follow the instructions in the NERSC folder to run the repository on the NERSC Perlmutter cluster.

Data

Please read the data folder for instructions on how to download/transfer/set up the data.