Datasets:
Tasks:
Fill-Mask
Sub-tasks:
masked-language-modeling
Languages:
English
Size:
10M<n<100M
ArXiv:
License:
Data encoding problem: text is stringified-bytes like "b'JEANNE D\xe2...'"
#10
by
yaakov
- opened
I added a parallel bug report on Github: https://github.com/Breakend/PileOfLaw/issues/17
The credit card agreements scraping process apparently produced binary data that is stored in text strings like "b'JEANNE D\xe2\x80\x99ARC CREDIT UNION\n...'".
Note that this is a str that contains the representation of binary data, not binary data itself (as binary data cannot be stored in json).
I noticed this in the first ten lines of the file data/train.cfpb_cc.jsonl.xz and suspect that it affects all the records in that file.