Three of the dataset files have been flagged as harmful.
Three of the dataset files have been flagged as harmful.
ru/ru_part_00211.parquet: Virus: Win.Trojan.URLspoof-1
zh/zh_part_00280.parquet: Virus: Win.Trojan.N-69
ru/ru_part_00639.parquet: Virus: Vbs.Worm.CoolNote-2
What have the author done to address this?
Hi @danieldux , regarding this issue, you can read more about it here:
- https://huggingface.co/datasets/allenai/c4/discussions/2
- https://huggingface.co/datasets/oscar-corpus/OSCAR-2201/discussions/12
In short, it seems that some documents in the OSCAR dataset contain content related to viruses, which causes HF to misidentify them. CulturaX is created by cleaning the OSCAR dataset, so it inherits this issue but with fewer files identified as viruses.
Hi @danieldux , regarding this issue, you can read more about it here:
- https://huggingface.co/datasets/allenai/c4/discussions/2
- https://huggingface.co/datasets/oscar-corpus/OSCAR-2201/discussions/12
In short, it seems that some documents in the OSCAR dataset contain content related to viruses, which causes HF to misidentify them. CulturaX is created by cleaning the OSCAR dataset, so it inherits this issue but with fewer files identified as viruses.
Hmm, interesting. I liked the suggestion from @uj to annotate spans of text that refer to malicious intent or code.