Datasets:
Question about data ordering/shuffling in the FinePDFs parquet files
#30
by
yujisw
- opened
Hi FinePDFs team,
Thank you for releasing this amazing dataset!
I am planning to use FinePDFs, and I have a quick question regarding the data ordering in the provided parquet files.
Is the data in the parquet files already globally shuffled, or is there any inherent bias in the ordering (e.g., sorted by crawl date, document length, or topic)?
I want to know if I need to perform a heavy global shuffle on my end before streaming the data to avoid bias.
Thanks in advance!
Best,
Yuji NARAKI
Hey, yes it's not shuffled and we would recommend shuffling