Triples in Trainer

[seplee]

Hi!

I'm trying to train answerai-colbert-small-v1 on data retrieved from a recommendation system, in this sort of scheme.

Thanks for posting an example on your README.md!

What are triples in Trainer?
I printed out a few characters of the commented MS MARCO file, but I am confused. What do these numbers mean?

Are they a mapping of the index of one collection to a list of queries and the relevance score of the queries?
ex: [indexed collection, [indexed query, relevance score] ... ]?

If so, do I need to index my collections and queries in order for Trainer to use them?

Thanks,
Lee