burn/examples/wgan
Guillaume Lagrange e718243748
Fix clippy errors w/ new rust stable (#3325)
2025-06-27 07:54:26 -04:00
..
examples Update to edition 2024 (#2931) 2025-03-20 10:28:51 -04:00
src Fix clippy errors w/ new rust stable (#3325) 2025-06-27 07:54:26 -04:00
Cargo.toml Chain lint inheritance [was: Disable new default clippy tests] (#3200) 2025-05-20 08:23:11 -04:00
README.md Update tch instructions (#2844) 2025-02-25 08:09:10 -05:00

README.md

Wasserstein Generative Adversarial Network

A burn implementation of examplar WGAN model to generate MNIST digits inspired by the PyTorch implementation. Please note that better performance maybe gained by adopting a convolution layer in some other models.

Usage

Training

# Cuda backend
cargo run --example wgan-mnist --release --features cuda

# Wgpu backend
cargo run --example wgan-mnist --release --features wgpu

# Tch GPU backend
export TORCH_CUDA_VERSION=cu124 # Set the cuda version
cargo run --example wgan-mnist --release --features tch-gpu

# Tch CPU backend
cargo run --example wgan-mnist --release --features tch-cpu

# NdArray backend (CPU)
cargo run --example wgan-mnist --release --features ndarray                # f32 - single thread
cargo run --example wgan-mnist --release --features ndarray-blas-openblas  # f32 - blas with openblas
cargo run --example wgan-mnist --release --features ndarray-blas-netlib    # f32 - blas with netlib

Generating

To generate a sample of images, you can use wgan-generate. The same feature flags are used to select a backend.

cargo run --example wgan-generate --release --features cuda