All experiments were performed on Red Hat Enterprise Linux 9.2 (Plow) (x86_64) with Intel(R) Xeon(R) Gold and NVIDIA A100 80GB PCIe. The Python version is 3.11.5. The PyTorch version is 2.0.1 and the CUDA used to build PyTorch is 11.7. For the first two experiments, we trained the flow models with CPUs only. For the third experiment, we used GPUs to accelerate the training of variational autoencoders.

Since our experiments contain many replicates and each replicate contains many flow models with different number of layers, it takes quite long to run. To reproduce our experiments, please consider using an HPC system. The command lines for running each experiment can be found in the corresponding folder.

