Skip to main content
If you have test case data stored in a CSV file, you can bulk-import it into an Adaline dataset instead of adding rows manually. This is the fastest way to populate a dataset with large volumes of test data for evaluation — including multimodal data like images and PDFs.

​
Import a CSV file

Open your dataset and click Upload CSV to start the import: Uploading a CSV file in Adaline

​
Import behavior

The system handles empty and existing datasets differently:
Dataset stateImport behavior
Empty datasetThe first row of the CSV becomes column names. All subsequent rows are added as test cases.
Existing columnsThe first row of the CSV must match existing column names exactly. Rows are appended to the dataset.
If your CSV looks correct but the import fails with a parsing error, try importing the CSV into Google Sheets first, then exporting it back as a CSV. Use the re-exported file in Adaline — this resolves most encoding and formatting issues invisible to the human eye.

​
CSV format and examples

The first row of your CSV is always the header row — these become (or must match) your dataset column names. Every subsequent row is a test case. Column names should match your prompt’s variable names exactly so they map correctly during evaluation.

​
Text-only CSV

A standard CSV where every cell contains plain text: 
user_question,context
What is the capital of France?,"France is a country in Western Europe."
Summarize this article.,"The article discusses the impact of AI on healthcare."

​
Text and images

To include images in a CSV, use a publicly accessible image URL or a base64-encoded image string as the cell value. Adaline automatically detects the modality. 
user_question,product_image
Describe this product.,https://example.com/images/product-a.png
What color is this item?,https://example.com/images/product-b.jpeg
Is this item suitable for outdoor use?,https://example.com/images/product-c.webp
Supported image formats: PNG, JPEG/JPG, GIF, and WebP.

​
Text, images, and PDFs

Combine all three modalities across different columns: 
user_question,reference_image,supporting_document
Describe what you see in the image.,https://example.com/images/storefront.png,https://example.com/docs/brand-guidelines.pdf
Compare this image to the brand guidelines.,https://example.com/images/new-design.jpeg,https://example.com/docs/brand-guidelines.pdf
Summarize the key findings from this report.,https://example.com/images/chart.png,https://example.com/docs/q4-report.pdf

​
Base64-encoded content

You can include base64-encoded images or PDFs directly in the CSV. This is useful when your files are not publicly hosted: 
user_question,product_image
Describe this product.,"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAA..."
What color is this item?,https://example.com/images/product-b.jpeg
You can freely mix URLs and base64 strings within the same column. Adaline handles each cell independently and detects the modality automatically.

​
How modality detection works

When Adaline parses a CSV, it inspects each cell value and determines the modality:
  • Plain text values are treated as text.
  • Values starting with http:// or https:// pointing to image files (.png, .jpg, .jpeg, .gif, .webp) are treated as images.
  • Values starting with http:// or https:// pointing to .pdf files are treated as PDFs.
  • Values starting with data:image/ are treated as base64-encoded images.
  • Values starting with data:application/pdf;base64, are treated as base64-encoded PDFs.
See Different Modalities in Dataset for the full guide on working with multimodal data in datasets, including size limits and storage behavior.

​
Next steps

Different Modalities

Work with text, images, and PDFs in dataset cells.

Setup Dataset

Create and configure datasets for evaluation.