Checkpoints | Dagnam.AI Docs

Checkpoints A running job saves checkpoints as it trains. This page covers how they are shown, how to download or restore one, and what deploying one looks like today. Best and latest The Checkpoints panel on a job's monitor page lists every checkpoint, newest first, with its epoch, step, size, how long ago it was saved, and up to four of its metrics. The checkpoint the job considers best carries a Best badge and sorts to the top. Checkpoints are not deleted individually; deleting a job from the Training jobs list removes its checkpoints along with it. Download \ndagnam checkpoint download checkpoint id ", , label: "Python", language: "python", code: "path = dagnam.download checkpoint \n job id,\n checkpoint id=None,\n prefer best=True, the best checkpoint instead of the latest\n ", , / On the web, click the download icon on a checkpoint's row. dagnam checkpoint list shows a job's checkpoints and their IDs. From the CLI or SDK, leave the checkpoint ID out to get the latest one, or pass prefer best=True in the SDK for the best one instead. Restore into a new job Click the restore icon on a checkpoint's row. A new job is created that resumes training from that checkpoint, and the web app opens it. With a Read & write personal API key you can also restore from the CLI, dagnam training restore , or the SDK, dagnam.restore checkpoint job id, checkpoint id . A restore is checked like any new job, so your plan's limits apply to it. Deploying a Studio model Deploying a checkpoint from a Studio-trained job is coming soon: the Deploy icon on a checkpoint's row is disabled. Today you can deploy a fine-tuned model version from My models . See Fine-tune a model /docs/training/fine-tune and Deployments overview /docs/deployments/overview .