Train

This chapter covers fine-tuning and pretraining, training runtime images, checkpointing and resuming TrainJobs and RayJobs, GPU slicing, and quota and scheduling governance for training jobs.

The components this chapter relies on are listed under Components below: Kueue handles queueing, quotas, and preemption, while Volcano and JobSet handle batch orchestration. Experiment tracking and model registration during training use MLflow from the Develop chapter.

Training Guides

Quota & Scheduling

Components