In preprocessing, how to divide samples to multiple tar files?

[Boltzmachine]

This is a follow-up of #48. I have collected betas grouped by image idx (3 for each image), but how do I split all data into multiple tar files as in the released dataset?

The original script https://github.com/MedARC-AI/MindEyeV2/blob/main/src/dataset_creation.ipynb splits by sessions, but it is not the case for mindeye 1 since all data should be collected first and then be grouped.