Sequence "packing" logic

by pietrolesci - opened Jan 5

EleutherAI org Jan 5

Hi there,

I am trying to understand how the tokenized Pile dataset has been "packed" into sequences of fixed length (2049 tokens). I started wondering about this when noticing (from a small sample) that the sequences are not concatenated using the <EOS> token, which is something I though was done by default.

Thanks a lot for your help!

pietrolesci changed discussion title from Data preparation implementation to Sequence "packing" logic Jan 5

pietrolesci

EleutherAI org Jan 9

Linking related GitHub issues: https://github.com/EleutherAI/pythia/issues/123#issuecomment-1882232326

markschoene

Oct 4

Hi are there any updates on this? The dataset has not been updated since and I can confirm that token id 0 (which relates to the EOD token) is not present in the data

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment