Datasets:
Tasks:
Automatic Speech Recognition
Languages:
Arabic
Multilinguality:
monolingual
Size Categories:
100K<n<1M
Language Creators:
crowdsourced
Annotations Creators:
expert-generated
Source Datasets:
original
Tags:
License:
Data files too large
#1
by
albertvillanova
HF staff
- opened
This discussion is a follow-up of this issue: https://github.com/huggingface/datasets-server/issues/1212
Currently the viewer shows a JobManagerCrashedError:
Job manager crashed while running this job (missing heartbeats).
Some of the data files are quite large (with more 1.5 GB).
One solution would be to split the data files into smaller ones (around maximum 500 MB each).
That was a hack to resolve the connection being dropped: https://github.com/huggingface/datasets/issues/4677
@msis please note this issue here is not related to the one you pointed out:
- The issue you pointed out is a 400 HTTP error generated by the
huggingface_hubclient, which is used under the hood when one pushes a local dataset to the Hub by usingpush_to_hub. - The issue here is that the operating system process crashed while trying to load the dataset (not pushing local files to the Hub), because the files are too large and do not fit in the RAM memory.