If you have a configuration file (any vissl compatible file) that you want to run on 1-gpu only (for example: train SimCLR on 1 gpu, etc), you don't need to modify the config file. VISSL also takes care of auto-scaling the Learning rate for various schedules (cosine, multistep, step etc.) if you have enabled the auto_scaling (see config.OPTIMIZER.param_schedulers.lr.auto_lr_scaling). You can simply achieve this by using the low_resource_1gpu_train_wrapper.sh script. An example usage:

cd $HOME/vissl
./dev/low_resource_1gpu_train_wrapper.sh config=test/integration_test/quick_swav

To run trainings (SimCLR, SwAV, PIRL etc.) for a few iterations, we provide run_quick_tests.sh. This requires 1 or 2 gpus.

Please see the script dev/launch_slurm.sh for running the code on slurm cluster for training purposes.

For every PR, we run/mandate a few checks before code is ready for review. The checks are:

1. flake8: We enforce coding style per file with defined flake8 rules. flake8 is a checker only and doesn't format code.
2. black: We format the code using black8
3. isort: We check that imports are properly sorted https://pypi.org/project/isort/. See setup.cfg for the settings.

In order to format code before code review, there are 2 options:

Run "./dev/linter.sh" at the project root before you commit. This will run isort, black and flake8 formatting.

We provide pre-commit hooks so as you build and commit (locally or github), the code formatting will be automatically run. You need to run pre-commit install once to enable this.

Read the doc https://ljvmiranda921.github.io/notebook/2018/06/21/precommits-using-black-and-flake8/ for how all the components operate and the pipeline/steps involved.