Currently, the corpus and queries are being transformed from the dataset to a dictionary. This process can be extremely slow for large datasets—for example, it took more than an hour to download MiracleRetrieval, with most of the time spent on type conversion to a dictionary