Let's Boogie
boogiebench evaluates models' creativity and musical and temporal reasoning.
Models are prompted to write a composition in strudel, a JavaScript music library. strudel offers a rich set of tools, including synths, samples, and effects. In this environment, LLMs should be able to express any tune they can imagine.
But how well can they imagine in music, having only been trained on text and images? boogiebench is here to find out.