Test run for a few good samples

[su-park]

In some evaluations, it can take a long time depending on the environment,
and errors are sometimes discovered late.
It would be good to be able to test a small number of cases to see if they work properly.

[KennethEnevoldsen]

This seems like a usage question so will transfer it over to discussions

[KennethEnevoldsen]

A solution might be to run the benchmark using a smaller model first. However many of the current datasets in MTEB are excessively large. We are working on a lite version of it, which should be a tremendous speed-up.