Large dataset is incomplete?

#1
by cckm - opened

Thanks for uploading this webdataset!

I noticed that the large dataset may be incomplete. Based on the libriheavy manifest (libriheavy is a strict subset of librilight) https://huggingface.co/datasets/pkufool/libriheavy, these files should be in librilight large, but they don't seem to be in this dataset under the *large*tar files (yes, I went through all of the *large*tar files and checked)
download/librilight/large/10205/grandther_baldwins_thanksgiving_1602_librivox_64kb_mp3/ballads_13_alger._64kb.flac
download/librilight/large/11969/rough_and_ready_1812_librivox_64kb_mp3/roughandready_01_alger,jr._64kb.flac
download/librilight/large/4344/ragged_trousered_philantropists_th_librivox_64kb_mp3/philanthropists_20_tressell_64kb.flac

Perhaps related to this, but I notice the *large*tar file numbers are not contiguous. For example,
librilight-large-wo6454-flac-000310.tar to librilight-large-wo6454-flac-000319.tar
librilight-large-wo6454-flac-000360.tar to librilight-large-wo6454-flac-000399.tar
does not exist.

Were some librilight files deliberately or accidentally omitted in this dataset?

Sign up or log in to comment