Source of training queries of BGE-EN-ICL (BEIR datasets)

[ftvalentini]

I have some questions regarding the origin of the training queries used for BGE-EN-ICL, which have no training queries in BEIR:

• quora: 10k test, 5k dev queries in beir -- bge-full-data has 60202 queries.
• scidocsrr: 1k test queries in beir -- bge-full-data has 12654 queries.
• arguana: 1406 test queries in beir -- bge-full-data has 3101 queries.

Where do these train queries come from?

Also for nli dataset: what is the source dataset?

Thank you so much for making such a valuable dataset available!

[545999961]

The dataset above is sourced from the reference paper in BEIR. The nli dataset is sourced from sentence-transformers/nli-for-simcse.