End to End Workflow with Base Command Platform and Fleet Command

Step 1. Create a Workspace

Start off by running

./1-create-a-workspace.sh

This creates a workspace on NGC and spins up a cpu job that will run "git clone" with the specified git repo.

Step 2. Upload Dataset

Next up, run

./2-upload-dataset.sh

This creates a training and validation dataset and uploads the files accordingly to NGC. In this example, we won't have a training dataset as we'll be using Tensorflow to pull the CIFAR-10 dataset directly.

Step 3. Start Training job

To kick off our training job run

./3-start-training-job.sh

This will spin up a job using NVIDIA's DGX GPU with a Tensorflow image. It will also mount the workspace and datasets we created earlier to be used for training and validation.

Step 4. Upload to the Private Registry

Once we've validated that the model is accurate run

./4-upload-to-ngc.sh

to upload the model to the private registry. If a model already exists the script will create a new version, but if the model doesn't exist it will create a new model and upload it as v1.

Step 5. Setup Fleet Command

Run the following to create a location and system on Fleet Command.

./5-setup-fleetcommand

Once this is setup, login to your system and connect it with the code generated from Fleet Command.

It will look like this: