To get setup just run pip install -r requirements.txt.
- Set an environment variable for the
WANDB_USER_NAMEto sync correctly w/ Wandb. - Set an envvar for the
DEVICE_TYPEto use non-cuda device. To launch training run: setup a config yaml file, then runpython -m train /path/to/config.yaml(ortorchrun).
Before commiting, we recommend running installing and running the pre-commit hooks:
pre-commit install # one-time command
pre-commit run --all-files- add your new model to
owl_vaes/models/{name}.py - add any new building blocks you need under
owl_vaes/nn - add your model to
owl_vaes/models/__init__.pyand give it an ID
- Review
owl_vaes/trainers/base.pyandowl_vaes/trainers/rec.pyto see general formatting - Implement your trainer in
owl_vaes/trainers/{name}.py - Add to
owl_vaes/trainers/__init__.pyand give it an ID
- See existing configs under
configsdirectory for examples.