I processed the datasets into several shards. If I want to load them as one piece I can do concatenation but it will take some time to index all of the files. Is there a quicker way to load the dataset like a memory mapping from several dataset shards?