This dataset is designed to evaluate embedding models for retrieval tasks, utilizing Wikipedia as the primary source.
By using Wikipedia as datasource, the dataset ensures that the content aligns with the knowledge domain of all embedding models.
The synthetic nature of the corpus guarantees that the specific sentences have never appeared on the web, eliminating concerns about data contamination.
