About the number of documents
Thanks for your efforts in collecting and cleaning the data.
I recently downloaded the full corpus from HF and found the # docs do not match with (and usually much less than) the provided ones in your repo. Just want to know if the statistics listed in the Table are for the unfiltered Common Crawl corpus?
Hello!
That's interesting. Can you provide the counts you got? (with the language you used, if possible)?
Thanks for your response. Here are some numbers for your reference:
| Language | # docs listed in HF | What I Got |
|---|---|---|
| Indonesian | 7,109,778 | 3,554,889 |
| Thai | 14,845,900 | 7,422,950 |
| Vietnamese | 33,933,994 | 16,779,803 |
I'll try to look at that in the following days. Thanks again :)
@lixin4ever Just out of curiosity, did you get this statistics by using the provided python script? Or did you download the data either through the Hub Python Library or Git?
I downloaded the data via Git and the numbers in the column "What I Got" are calculated by myself.
Any updates ?
In french the numbers do not seem to match either...
79167499 rows in the dataset (half as advertised ?) Was there subsequent filtering ?
Size is 470 Gb, close to what is stated.
Thanks for the great work on Oscar !