Jobs
The Jobs section is the full history of your training runs. It mirrors client.jobs in the SDK. Each row shows the job ID, type, status, assigned hardware, and when it was created.
Job status
| Status | Meaning |
|---|---|
queued |
Waiting for a GPU to be assigned. |
running |
Actively training on a GPU. |
completed |
Finished successfully; an output model is available. |
failed |
Stopped with an error (shown in the job detail view). |
cancelled |
Stopped manually before completion. |
Running jobs show a pulsing indicator and also appear on the Overview page under "Activity on the GPUs".
Watch a job
Each job row has quick actions:
- Metrics: opens a live chart of loss and learning rate across steps, with a trendline toggle to see through the noise. Metrics keep updating while the job runs and stay available after it finishes.
- Logs: streams the job's console output live.
- View: the job detail panel: full ID, status, assigned GPU (and the requested tier if it differed), VRAM and cost estimates, and any error.
Cancel a job
For any queued or running job, click Kill. You'll be asked to confirm. The GPU is released immediately, and any progress not saved to a checkpoint is lost. A cancelled job produces no output model.
After completion
When a job completes, its fine-tuned output appears in Models under "Fine-tuned", ready to deploy. Intermediate checkpoints are pruned automatically per the checkpoint limit. See Training and Storage.