ahaeflig/vae_bench

Variational Auto Encoder Simple Benchmarks

In this repository I plan to benchmark VAE variants focusing on reproducibility and ease of reading. Goals:

1. One file per model
2. Run all experiments with one command
3. Easy experimentation

Installation (Linux + Docker + Nvidia GPU)

1. Install the NVIDIA drivers.

2. Install Docker ce.

3. Manage docker as a non-root user.

4. Install nvidia-docker.

5. Clone or copy this repository, assuming it was copied in $WORKDIR

6. Run the installer

cd $WORKDIR/
sudo chmod +x install/install.sh
./install/install.sh

Run the benchmark script

cd $WORKDIR/
docker run --rm --gpus all -v $PWD:/home/vae/project/ vae:latest python3 benchmark.py