BYOT: Bring Your Own Training

This repo acts as a template repo to bring your own custom training based of this feature guidebooks/store#212

How do I get started

You can clone this template and extend this and launch your own custom training.

We are running this command behind the scenes, here CUSTOM_WORKING_DIR will refer to your directory that was cloned off this template.

ray job submit --job-id ${JOB_ID} --no-wait --runtime-env ${CUSTOM_WORKING_DIR}/runtime-env.yaml --working-dir ${CUSTOM_WORKING_DIR} --address ${RAY_ADDRESS} -- python main.py

Broadly, these are the main components of this repo:

1. main.py acts as the entry point and this is what will be executed when ray job submit will be executed.

2. runtime-env.yaml will act as a a way to provide additional dependencies, and will be installed because of this param in the above command --runtime-env ${CUSTOM_WORKING_DIR}/runtime-env.yaml

3. src/ can act as the placeholder for the rest of your python code. or you can create an other additional directories here. --working-dir ${CUSTOM_WORKING_DIR} flag in the above command makes sure that we copy the entire working directory over to the CodeFlare cluster, so all files under your directory will be available to you.