Train your Assistant

This guide will walk you through the process of training your assistant in Rasa Studio. By the end, you'll understand how to initiate training, monitor its progress, resolve common errors, and debug issues.

Before Starting

Training a model requires specific permissions that are not available to every Studio role. Ensure your role has the right access before continuing.

About Training

Training compiles your flows, NLU data, and other inputs into a functional model, that you can use for testing or deploy to production.

Rasa Studio provides two ways to train your assistant:

⭐️ Training in Studio (Recommended):
For rapid iteration and testing your assistant within the Studio UI.

Training your assistant locally with Rasa Pro:
Useful for technical users who have additional custom components or advanced integration steps that they want incorporated before training.

note

If you're interested in diving into technical details of how training works see more in our Training Architecture Reference.

Training in Studio

Step One: Get Started

1. Access Training Navigate to the Flow builder, NLU, or CMS pages. You should be able to see the "Train" button:

2. Prepare Your Flows In the Flow list, toggle the "Ready for Training" switch for all flows you want to include. Only flows marked as "Ready for Training" will be included in the model.

3. Initiate Training Once your flows are ready, click the "Train" button. Studio will validate your flows before beginning the training process.

Step Two: Validate and Improve

Rasa Studio runs validation your model before the training process begins. This step helps you catch common issues—like missing intents or invalid syntax—early in the process, saving time and effort.

1. Resolve Validation Errors If errors are found during validation, you’ll see an "Unable to train" message.

   • Click on the flow names in the error list to locate and fix issues.
   • Examples of errors include missing intents, invalid syntax, or unsupported actions.

2. Monitor Progress After validation, Studio will display a training progress message. Other users can view the status but cannot stop or restart training until it’s complete.

How to Manage Training Outcomes

Once validation is complete, training begins automatically. Studio provides details on the outcomes as well as logs to help you debug any issues that might come up.

1. Handle Training Outcomes

   • Successful Training: A green checkmark and success message indicate the model is ready.

   • Training Failure: If training fails, download the logs from the Versions page to investigate. Typical causes include invalid training data or conflicting rules.

2. Stop Training (Optional) If necessary, the user who initiated training can stop it. This will terminate the process and display a status update.

Troubleshooting Tips

1. Access Logs and Versions Go to the Versions page to view trained versions of your assistant. Use the ellipsis (three dots) next to a version to download the model or training logs for further analysis.

2. Reproduce Errors Locally

   • Download the model input files from Studio.
   • Set up a matching local Rasa Pro environment.
   • Run:

     rasa train --debug

   This will provide detailed error messages to help identify the issue.

Training Your Assistant Locally using Rasa Pro

Step One: Download your Assistant Files

Option 1: Downloading via the Studio UI

• Navigate to the Versions page in Studio, locate the version you want to train, and click the ellipsis (three dots) to download the assistant files.

Option 2: Downloading via the Rasa Pro CLI

Replace [project-name] with the name of your project.

rasa pull --project [project-name]

Step Two: Train Your Assistant

Place your assistant files in the working directory and run the following command to start training:

rasa train

For more detailed local training steps using the Rasa Pro you can take a look at the technical reference to train your assistant locally.