Create a Workflow

📍

Step 2 of 3: Create a Workflow

Part of the training workflow quickstart. Next: Start a training run.

Configure a reusable training workflow using a visual node-based interface. Workflows define your model, dataset split, and training parameters.

⏱️ Time: ~5 minutes

💡

Need detailed guidance?

This is a quickstart overview. For comprehensive configuration:

• Complete workflow guide
• Configure your dataset
• Configure your model
• Training settings

---

Create your workflow

From your training project overview, click Create Workflow

The workflow canvas opens with a node-based interface. You'll configure three components from top to bottom: System Prompt, Dataset, and Model.

---

1. Configure system prompt

Click the System Prompt node at the top. This defines instructions for your VLM.

📘

For quickstart

Keep the DEFAULT prompt. You can customize this for specific tasks in future trainings.

---

2. Configure your dataset

Click the Dataset node to select your data source and split strategy.

1. Dataset Source — Select the dataset you created earlier

2. Keep the default split ratio and shuffle settings for your first training

Learn more about dataset configuration →

---

3. Select and configure your model

Click the model dropdown to choose from available VLM architectures.

📘

Quickstart recommendation

Select any available model and keep the default settings. You can experiment with different models and parameters in subsequent trainings.

After selecting a model, click the model node to view its configuration.

Keep all default parameters for your first training run.

Learn about model settings →

---

Review and run

Once configured, your workflow canvas shows all connected nodes.

Click Run Training to proceed.

✅

Workflow configured!

Your workflow is ready. Next, start your first training run.

---

Common questions

Can I reuse this workflow?

Yes! Workflows are reusable. You can run the same workflow multiple times or use it with different datasets.

What if I want to change settings later?

You can edit the workflow nodes anytime before running, or create a new workflow with different settings.

Which model should I choose?

Any available model works for quickstart. Keep default settings for your first training, then experiment with different models in subsequent runs.

Compare model architectures →

---

What's next?

Start a Training Run →

Launch training and monitor your VLM's progress in real-time.