Trying to test our systems document summarizer using promptfoo. Our documents are fairly large. It feels like to do a good test I'd need to have the chunks and use those as input, but I might also need to chunk up the known good summary I want to compare against due to token limitations? Any guidance around what best practices would be for this?