Graph Convolutional Network in MLX

An example of GCN implementation with MLX. Other examples are available here.

The actual benchmark on M1 Pro, M2 Ultra, M3 Max and Tesla V100s is explained in this Medium article.

Install env and requirements

CONDA_SUBDIR=osx-arm64 conda create -n mlx python=3.10 numpy pytorch scipy requests -c conda-forge

conda activate mlx
pip install mlx

Run

To try the model, just run the main.py file. This will download the Cora dataset, run the training and testing. The actual MLX code is located in main.py, whereas the PyTorch equivalent is in main_torch.py.

python main.py

Run benchmark

To run the benchmark on CUDA device, a new env needs to be set up without the CONDA_SUBDIR=osx-arm64 prefix, to be in i386 mode and not arm. For all other experiments on arm and Apple Silicon, just use the env created previously.

python benchmark.py --experiment=[ mlx | torch_mps | torch_cpu | torch_cuda ]

Process benchmark figure

This needs to install additional packages: matplotlib and scikit-learn.

python viz.py