Use case
Bring your training script and data to a GPU workbook.
Use the GPU training starter to verify the runtime, stream loss metrics, save a checkpoint, then deliberately replace the example with the data and training code your project requires.
Training is a workload you own, not an opaque hosted pipeline. Keep the scripts, dataset references, checkpoints, metrics, and the agent's changes together so a later run is understandable and repeatable.
- GPU training starter with live loss metrics
- Persistent checkpoints and saved reports
- Room for your own training script and authorized data
- Explicit runtime limits and cost estimates