How to inspect training data in a particular batch?

[explanare]

❓ The question

I would like to inspect training data in a particular batch as described here. I also noticed that there is another script inspect_train_data.py doing similar things. Unfortunately, both scripts require the global_indices.npy file, which is no longer available on the server, e.g., for the OLMo-7B model, https://olmo-checkpoints.org/ai2-llm/olmo-medium/wvc30anm/train_data/global_indices.npy is no longer accessible.

Is there a new URL where we could download these global_indices.npy files for the pre-trained models, or do we have to rebuild the global_indices.npy locally? Also, are there any tools to sample training data for a particular batch without downloading the entire training corpus?

Thanks in advance!