physics-vqvae

Using Vector-Quantized VAEs to model physics

Installation

conda create -n g_vqvae python=3.12
conda activate g_vqvae
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128
pip install torch_geometric
pip install einops h5py imageio ipykernel matplotlib pandas wandb dotenv prodigyopt
pip install scipy pytest
pip install -e .

Usage

To train the VQVAE on a dataset, run the training script. Before this, make sure you have a dataset in correct folder. Additionally, wandb requires an api key to be set, we assume it is placed in a .env file in the root directory.

WANDB_API_KEY=xxx