Train a Model
Create a training project, configure a workflow, and start a training run to fine-tune your VLM in Datature Vi.
- At least 20 annotated images
- Compute Credits available in your account
After completing this stage, you will have a training project with a configured workflow and an active or completed training run.
Datature Vi organizes training into projects and workflows. A project is a container for all your training work on a given task. A workflow is a reusable configuration that specifies your model architecture, dataset, and training parameters. When you start a run, Vi launches training on GPU infrastructure using that configuration.
This stage has three steps.
Create a training project
Click Training in the left sidebar, then click Create Training Project. Give your project a name, add an optional description, then pick localization: Multi-Region is the default; use a single region if your org needs fixed geography for data (Security and compliance). Click Next, review the summary, and click Create Project.
What each step covers
Create a training project
Set up the project container with a name and storage localization.
Create a workflow
Connect a system prompt, a dataset, and a model on the visual workflow canvas.
Start a training run
Configure hardware, pass dataset validation, and launch training. Runs typically take 1 to 3 hours depending on dataset size, model, and GPU type.
Next steps
Updated 30 days ago