Datasets:
is ` ultrafineweb_l1_en_hq` a replacement for the old ultra-fineweb splits, or should it be used in conjunction with them
question is in the title ^~^ thanks!
Our suggestion is to use them in conjunction with ultrafineweb_l1_en-hq, which is based on our latest open-source Ultra-FineWeb-L1 and uses the same Ultra-FineWeb classifier for filtering. Details: https://huggingface.co/datasets/openbmb/Ultra-FineWeb-L1
Built from Common Crawl snapshots, it undergoes main-text extraction, language filtering, heuristic filtering, sensitive-field replacement, customized cleaning, and deduplication, yielding 1T+ tokens (approximately 1.14 billion documents). We simultaneously release the L2 selected data Ultra-FineWeb, selected by the Ultra-FineWeb classifier. To our best knowledge, this open-source web pre-training dataset covers the most recent Common Crawl snapshots, up to CC-MAIN-2025-51. πππ
