Some benchmarks have a lot of duplicates, so we compress the data by
aggregating them, the number of duplicates for each benchmark is then
stored.