Skip to main content

Exercise 2 - Swap the Generator

Goal: Introduce the Generator abstraction from the theory and plug Gemini into the pipeline you built in 02-generation with a single line change.

Background​

Right now your pipeline calls Ollama directly. If you want to switch to Gemini, you have to find and edit every place the model is called. The abstraction fixes this: the pipeline accepts a Generator and does not know or care which model is underneath.

Assignment​

Open 02_swap_generator.py.

  1. Define a Generator base class with a single generate(prompt: str) -> str method.
  2. Implement OllamaGenerator and GeminiGenerator subclasses.
  3. Copy your ask(question, index, chunks, generator, k) function from the previous module - change only the signature to accept a generator instance instead of calling Ollama directly.
  4. Run the same five questions from Generation Exercise 2 using both generators.
  5. Compare the answers side by side. Are there questions where one model is noticeably better?

Thinking questions​

  • Your ask function now has no import requests inside it - all model-specific code is in the generator classes. What are the testing benefits of this design?
  • If you wanted to add a third generator (e.g. OpenAI GPT-4), how many lines of the pipeline would you need to change?
  • Gemini has a 1 million token context window. Does that mean you should always pass more chunks? What is the argument against it?

← Exercise 1