Add Parallel Coordinates plot option when there are multiple eval metrics

Suppose we want to choose the ‘best’ prompt template p among N in the sense that it maximizes metrics m0, m1, … mM. We’re run every template through with multiple parameters, but we only care about the template as a parameter. Then, we want to quickly visualize this data across metrics m_i.

For a single LLM case and a reasonable num of prompt templates (eg <20), we can use a Parallel Coordinates plot: https://www.data-to-viz.com/graph/parallel.html

The N prompt templates are different colors on the plot, where each prompt template p has R lines (number of responses in that template).

For multiple LLMs, we could lower opacity and let users select one ‘slice’ of the whole plot to view which visualizes a single LLM. Scrolling through the parallel coord plots then gives us a picture of what’s going on.