Completion of custom model training job – Vertex AI Workbench and Custom Model Training

Based on the complexity of the model, model training might take few mins to few hours. Once the model training is completed it will be listed under the training section and also the model will be available in the model section.

Training job will be listed under training section of vertex AI as shown in Figure 4.33:

Figure 4.33: Training module containing training jobs

1. Select Training module under Vertex AI.
2. Training job will be listed under model type to be custom.

Trained model will be listed under model registry of vertex AI as shown in Figure 4.34:

Figure 4.34: Model registry containing trained model

1. Select Model registry of Vertex AI.
2. Trained model will be listed under custom trained type.

Model will be exported along with the artifacts to the cloud storage bucket as shown in Figure 4.35 (bucket details were mentioned in the Python code):

Figure 4.35: Cloud storage containing trained model and artifacts

Deletion of resources

All the compute resources are expensive. Follow the steps as shown in Figure 4.36 to delete the notebook instance. Follow the steps in the previous chapter to delete the model from the model registry, training pipeline and model stored in the cloud storage:

Figure 4.36: Deletion of notebook instance

1. Select the notebook name.
2. Click DELETE (A pop up will appear for users confirmation).

Conclusion

In this chapter we learnt about workbench in detail with various components of it. We have covered how we can create python files in the workbench, how to create image using Dockerfile, how to submit custom model training using vertex AI of GCP.

In the next chapter we will start understanding how hyperparameter tuning can be performed for the custom models.

Questions

1. What is the difference between managed notebooks and user managed notebooks?
2. Why is service account needed while submitting custom training job?
3. Can users take the custom trained models for deployment?