Context Engineering with DSPy

Chapter 6. Deep Dive into Prompt Optimizers

A note for Early Release readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 6th chapter of the final book. The GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Reaching the optimization stage means you’ve done your homework. You’ve defined your program, collected your dataset, and built a robust evaluation metric. Your reward is that you now have everything you need to run DSPy’s powerful optimizers. But it can be hard to understand which one to run.

Different optimizers in DSPy will improve the quality of your program’s outputs by selecting or generating good few-shot examples, or proposing and testing better prompt instructions, as well as fine-tuning the weights of your language model. Finding the right optimizer and configuration to use for your task requires experimentation, but there are good rules of thumb to follow:

• Start with BootstrapFewShot if you have ~10 examples in your dataset

• For optimizing the prompt, GEPA learns well from 10-50 examples

• If you have 50+ examples try BootstrapFewShotWithRandomSearch

• If your optimization budget is substantial and you have 200+ examples, use MIPROv2

• To maximize efficiency ...