Skip to content

Models Tests

Note

More tests to come soon...

🌺🍕 Hawaiian test

The Hawaiian Test is a simple, practical technique invented by Philippe Charrière to verify the proper functioning of a generative AI setup—specifically, to ensure your selected model actually follows instructions and deliver expected responses.

Purpose

  • To avoid wasted effort when building AI tools where the model ignores your instructions.
  • To quickly check if system messages and prompt engineering work as expected for your model.

How the test works

It involves five main checks: Does the model itself respond as expected?

  1. who are you?
  2. who invented Hawaiian pizza?
  3. what are the ingredients of a Hawaiian pizza?
  4. what are the regional variations of Hawaiian pizza?
  5. what is the best pizza?

Why "Hawaiian"?

The name comes from the test prompt: you instruct the model (via a detailed system message) to behave as a Hawaiian pizza expert. You then ask questions about Hawaiian pizza. If the model answers as a knowledgeable, enthusiastic Hawaiian pizza expert and follows your constraints (focus on Hawaiian, correct history, detailed ingredients, etc.), the setup passes the test.

Typical "Hawaiian Test" steps

  1. Send a detailed system prompt: e.g., “You are Bob, a Hawaiian pizza expert. Provide enthusiastic, accurate info about history, ingredients, and regional varieties...”
  2. Ask targeted questions: “Who are you?” “Who invented Hawaiian pizza?” “What are the ingredients?” “What are the regional variations?” “What is the best pizza?”
  3. Evaluate responses: The AI should reply in character, only use provided knowledge, defend pineapple on pizza, and keep focus on Hawaiian pizza even when asked about other pizzas.
  4. Success: If the responses match your expectations, you can trust this stack for further development. If not, you know where troubleshooting is needed (API, model, or framework).

In short, the Hawaiian Test is a fast, domain-specific way to validate prompt control and model behavior for LLM projects—using a whimsical but demanding scenario as the benchmark.[Ref1]

Info

The Hawaiian Test Applied to Pydantic AI and Docker Model Runner: Ref1