tg fine-tuning retrieve <JOB_ID> and tg fine-tuning list-events <JOB_ID>, or open the jobs dashboard.
Job timing
Time to start: A job moves throughpending, queued, and running before training begins. Start time depends on queue depth and hardware availability. With no other pending jobs and free hardware, a job usually reaches running within a minute, and most jobs start within an hour. There is no SLA on queue wait time.
Run time: Run time depends on model size, dataset size, and network speed during the model and data downloads. Once a job is running, its estimated time remaining may appear on the jobs dashboard or in the tg fine-tuning retrieve output.
Diagnose a failed or stopped job
A job that doesn’t finish ends in one of three states:- Cancelled: Usually an automatic cancellation, most often from a failed balance check. Add funds or raise your spending limit, then resubmit.
- User error: A problem with your inputs, most often a dataset that fails server-side validation or an incompatible continued-fine-tuning checkpoint. The event log names the specific reason. See Data validation failures below.
- Error: A problem on Together AI’s side. These are usually transient. Our support team will follow up about the failure, or you can reach out at together.ai/contact.
File upload failures
If a training file fails to upload, the CLI and web UI return the specific reason. A 401 or 403 status means your API key is missing or inactive, so verify it’s set and active. Other failures, such as an insufficient balance, are described in the returned error.Data validation failures
Most dataset validation runs server-side during ingestion, after upload and before training. If a job stops with a data-related error, inspect the event log and the file’s validation report. See Wait for server-side validation for how to readprocessing_status and validation_report.