This dataset is a random sample of ArXiv abstracts labeled as astro-ph. It is intended for training vector similarity models.
It also has a BEIR-compatible version of the test split that can be used to measure the accuracy of vector models trained with this data.
