Training Models with AutoTrain

HuggingFace AutoTrain lets you train custom ML models without writing training code. XGENIA integrates AutoTrain through AI tools and visual pro nodes.

What you will learn in this guide

• How to create an AutoTrain project
• How to upload training data
• How to start and monitor training
• How to scaffold a complete ML pipeline on the canvas

Supported Tasks

Task	Description	Example Use Case
text-classification	Classify text into categories	Sentiment analysis, spam detection
tabular-classification	Classify tabular data rows	Churn prediction, fraud detection
tabular-regression	Predict numeric values from tabular data	Price forecasting, scoring
image-classification	Classify images into categories	Product categorization
llm-finetuning	Fine-tune a large language model	Custom chatbots, domain-specific AI
dreambooth	Fine-tune image generation models	Custom image generation

Training via AI Chat

Step 1: Create a project

Create an AutoTrain project called "sentiment-model" for text classification

The AI calls autotrain_create_project and returns a project ID.

Step 2: Upload your data

Upload training data to the sentiment-model project from https://example.com/train.csv

The AI calls autotrain_upload_data with the project ID and data URL.

Step 3: Start training

Start training the sentiment-model project

The AI calls autotrain_start_training. Training runs on HuggingFace infrastructure.

Step 4: Check status

What's the status of my sentiment-model training?

The AI calls autotrain_status and reports progress, including any trained model repositories.

Scaffolding a Full Pipeline

Ask the AI to set up a complete ML pipeline on your canvas:

Create an AutoTrain workflow for text classification

This calls create_autotrain_workflow and places the following nodes on your canvas:

1. HF Token variable — stores your API key
2. Project Name variable — the AutoTrain project name
3. Dataset URL variable — where your training data lives
4. Create Project — JS function node that creates the project
5. Upload Data — JS function node that uploads the dataset
6. Start Training — JS function node that triggers training

All nodes are pre-wired using @label references. Just set your token, dataset URL, and trigger the pipeline.

Using the Auto ML Trainer Pro Node

For more control, use the Auto ML Trainer visual node.

Inputs

Port	Type	Description
data	Array	Training data (array of records)
targetColumn	String	Column to predict
taskType	Enum	text-classification, tabular-classification, etc.
baseModel	String	Base model from HuggingFace Hub (optional)
hyperparameters	Object	{ learning_rate, num_epochs, batch_size }
hfToken	String	HuggingFace API token
mlServerUrl	String	ML Coordinator URL
Train	Signal	Start training
checkStatus	Signal	Poll for training progress

Outputs

Port	Type	Description
status	String	Current training status
hfProjectId	String	AutoTrain project ID
hfModelRepo	String	Trained model repository on HuggingFace
trainingMetrics	Object	Accuracy, loss, and other metrics
error	String	Error message if training fails
statusChanged	Signal	Fires when status changes

Typical Workflow

1. Wire your analyzed data and target column from the Auto ML Analyzer
2. Set the taskType matching hfTaskType from the analyzer
3. Trigger Train to start
4. Poll with checkStatus or listen for statusChanged
5. When complete, use hfModelRepo with the Auto ML Predictor node