Train a Model

📍

Quickstart: Step 2 of 3

This is the second step in the quickstart guide. You should have already prepared your dataset. After training, you'll deploy and test your model.

Now that your dataset is ready, it's time to train your VLM. Datature Vi makes training straightforward with organized training projects and reusable workflows.

⏱️ Time to complete: ~10 minutes, excluding training time

📚 What you'll learn: Training project setup, workflow configuration, and running training

📋

Prerequisites

• A dataset with images and annotations (complete step 1 if you haven't)
• At least 20 annotated images for training

---

Three steps to train your model

Follow these steps in order to train your VLM:

---

Quick overview

What you'll do

1. Create a training project — Organize your training work in a dedicated project
2. Create a workflow — Define model architecture, dataset split, and training parameters
3. Start a training run — Launch training and track metrics

What you'll need

• A dataset with annotations (from step 1)
• About 10 minutes for setup
• Training will run in the background (typically 1-3 hours depending on dataset size)

💡

Need more detail?

This quickstart covers the essentials. For comprehensive guides, see:

• Complete training project guide
• Complete workflow configuration guide
• Training settings documentation

Tips for success

• Start with default model settings for your first training run
• Use an 80/20 train/test split for initial experiments
• Monitor training metrics to identify overfitting
• Save checkpoints regularly to preserve progress
• Choose Multi-Region localization for best availability

---

Understanding training workflows

Workflows are reusable training configurations that define:

• Model architecture — Which VLM backbone to use
• Dataset configuration — Train/validation/test split
• Training parameters — Learning rate, batch size, epochs
• GPU resources — Compute requirements

Once created, workflows can be reused for multiple training runs with different datasets or settings.

---

What happens during training?

When you start a training run:

1. Initialization — Model loads and dataset prepares
2. Training — Model learns from your annotations
3. Validation — Performance evaluated on validation set
4. Checkpointing — Model state saved at intervals
5. Completion — Final model ready for evaluation and deployment

You can monitor progress in real-time through the training dashboard, which shows:

• Loss curves
• Validation metrics
• Training logs
• Resource utilization

---

What's next?

Once your model finishes training, you'll evaluate its performance and deploy it for inference.

Deploy and Test Your Model →

Download your trained model and test it with new images to see how well it performs.