Create a fine-tuning job

This guide walks you through creating a fine-tuning job using the Qwen Cloud console.

​

Prerequisites

• A Qwen Cloud account. Log in to the console.
• A published training dataset in JSONL format. See Create a dataset to upload your data, and Datasets overview for format requirements.

​

Create a job

Navigate to the Fine-tuning page in the console and click Create fine-tuning job.

​

1. Select model and method

• Base model: Choose from the dropdown (e.g., Qwen3-14B). Switch to the Custom Models tab to use a previously fine-tuned model for iterative training.
• Job name: Enter a name or leave blank to auto-generate one.

​

2. Select training data

Click Select training datasets to choose a published dataset from your Datasets library.

​

3. Configure validation (optional)

Optionally monitor training quality with validation:

• Auto Split: Automatically split a portion of training data for validation. Adjust the ratio with the slider.
• Custom Dataset: Select a separate dataset for validation.

​

4. Configure output model

• Model name: Enter a name for the fine-tuned model.
• Export quantity limit: Set the maximum number of checkpoints to save.
• Checkpoint save interval: Save by epoch or by step, and set the save frequency.

​

5. Review cost and submit

Review the estimated cost based on your selected model and dataset size, then click Create fine-tune job to submit.

​

After creation

Once submitted, the job progresses through the following statuses:

1. Pending -- Job is submitted and waiting for resources.
2. Queued -- Job is in the scheduling queue.
3. Running -- Training is in progress. Monitor metrics on the job detail page.
4. Completed -- Training finished successfully. The fine-tuned model is ready.

If training encounters an error, the status changes to Failed. Check the job logs for details.