Welcome to this quick guide on getting started with Arato and the basics of how to run your first experiment.
Before running your first experiment, we recommend reading about Arato Notebooks.
Step 1: Creating a New Notebook
A notebook in Arato acts as your workspace where you can organize, build, test, iterate, and compare changes to your prompts, models, or configurations. To create one:
Click on "New Notebook"
At the top of the Notebook, you can give it a name to help keep your work organized. For this guide, we'll call it "Getting Started with Arato."
Step 2: Creating Your First Prompt
A prompt defines the instructions you want the model to follow:
Enter your prompt text in the User section of the prompt editor.
You can optionally edit your prompt title and give it a meaningful name. This will help you keep things organized.
Step 3: Defining Variables (Dynamic Inputs) in Your Prompt
Variables allow for dynamic generation of prompts, making it possible to test an LLM with multiple inputs efficiently. To define a variable:
Use double curly braces {{var_name}} to define a variable; the variable name should not contain spaces or special characters.
The text should turn blue, indicating it's recognized as a variable.
For example, let's define a variable named {{country}} in our prompt:
Your prompt now functions as a prompt template.
Step 4: Connecting Your Data to the Experiment
To run an experiment with your own data, you need to add a dataset to your prompt:
If you added variables to your prompt, it will automatically create a dataset that you can manually fill in.
Alternatively, you can upload a CSV file containing the necessary data.
Ensure that column names in the CSV match the variable names in your prompt.
You can optionally name your dataset by clicking its header.
Step 5: Setting Up the Model
The next step is setting up the model:
Click on the model dropdown and select your preferred vendor and model. Arato supports all models.
Enter your API token and click "Save" to confirm your selection.
Step 6: Running Your First Experiment
Once your dataset is ready:
Click "Run" to execute your first experiment.
View the Run Results to see an overview of variables and responses.
Click on a Run Result line to access the response details.
Step 7: Iterate & Expand
After your first run, you can now start experimenting:
Make changes to your prompt based on initial results
Use different input data in your dataset
Try different models to compare performance
Now you can collaborate with your teammates - invite team members to your notebook to share results and build on successful experiments together.
Don't worry - all changes are tracked for reference.