Why are inspirations passed to top_programs during prompt construction?

[kongwanbianjinyu]

I noticed that in controller.py, the prompt construction passes the inspirations list to top_programs. However, top_programs is expected to represent the top-k best-performing programs based on the evaluation metric, whereas inspirations may include programs selected for diversity or novelty and not necessarily those with the highest scores.

This raises a potential inconsistency: why are we treating inspirations as top_programs? Shouldn’t the prompt construction rely on the actual top-k scored programs instead?

[codelion]

Inspirations are probably useful for showing the LLM different options of implementation but we can refactor it to be separate from top-k programs.