Configure training

Before creating a run

Have a published dataset version and a compatible base model. Keep separate evaluation data. Check the project’s training-readiness messages, account permissions and any compute/credit requirements.

Configure the run

1. Open training from the intended project.
2. Select the exact dataset version. Check the task type and class information.
3. Select a registered base model compatible with the CV or VLM workflow.
4. Review the parameters exposed by the form and the compute destination. Start with the platform’s applicable defaults; change parameters for a documented reason.
5. Check any cost or approval information before submitting once.

A local path on your laptop is not a model registered on the cloud service. If a model is absent from the selector, ask about registration or the supported upload workflow.

Follow the job

Watch the task’s status, progress and available logs. A queued job is waiting; a successful submission is not a completed training run. Check the terminal status and saved output before reporting success.

If a request times out, look for the existing task before resubmitting. If it fails, record the task ID and error. Do not repeatedly retry a configuration that fails the same readiness check.

Compute and billing

Availability depends on the account and configured workers. This guide does not promise immediate GPU capacity, a free quota, a particular training duration or support for every model architecture. Confirm those conditions in the platform or with Otterview.

Next: evaluate model outputs.