codingchild2424 / vision

Clean, reproducible, boilerplate-free deep learning project template.

PyTorch Project Template

This is an example of a clean, reproducible, and boilerplate-free deep learning project that was developed as part of the Groomata deep learning course.

This project is organized using pytorch-lightning, and all configurations and artifacts can be uploaded to wandb without any compromise. You can see an example wandb workspace here. All configurations are programmatically generated and maintained by hydra and hydra-zen.

Usage example

docker run \
    --gpus=all \
    --ipc=host \
    --volume=/path/to/volume:/vision/.cache \
    --env-file=/path/to/.env \
    --tty \
    groomata/vision \
    # Override any configurations you want
    optimizer.lr=0.0001 \
    datamodule.dataloader.batch_size=64 \
    trainer.max_epochs=100 \
    trainer.gradient_clip_algorithm="norm" \
    trainer.gradient_clip_val=1.0