Training Jobs
Distributed training runs, the workhorse for serious model work.
Training jobs run finite, distributed training workloads on Vantage. They're built on Kubeflow Trainer under the hood, but you don't see any of that, you pick a runtime, a sizing, optional initializers for dataset and model, and where to put the output.
Runtimes
A runtime is a pre-built training environment: framework + base image + parallelism strategy. Vantage ships several cluster-wide runtimes; admins can publish workspace-scoped ones too.
Presets
Training presets are read-only named configurations that combine a runtime with a sizing, they let you start a job in a few clicks instead of filling out the full form. See Presets.
Next steps
- Presets: quick-start configurations
- Runtimes: pre-built training environments
- Submitting a job: the job creation form
- Lifecycle: phases and actions
- Reference: every field, every default