FineWeb2-HQ-Plus

Dataset summary

FineWeb2-HQ-Plus is a high-quality, model-filtered pretraining dataset derived as a subset of FineWeb2(v2.1.1), spanning 100 languages. It is an improvement over FineWeb2-HQ and was created by selecting the top 10% quality documents of FineWeb2 in each language, using scores assigned by our quality classifier based on XLM-RoBERTa embeddings.

Compared to FineWeb2-HQ, we changed the classifier recipe in two ways. First, we train a single classifier jointly on all languages from a pool of positive samples with additional multilingual instruction data (including MURI-IT, Tagengo, and WikiQA), with lower-resource languages upsampled by at most a factor of three so that high-resource languages do not dominate the decision boundary. Second, we draw the negative samples from the 50th–75th percentile of a proxy quality classifier rather than uniformly from FineWeb2.

For more details, see our paper Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection.

Key features

• High-quality selection: Top 10% of FineWeb2 documents by quality
• Multilingual coverage: 100 languages, ensuring diverse linguistic representation
• Model-based filtering: Uses an XLM-RoBERTa embedding-based classifier to score documents
• Enhanced benchmark performance: Surpasses FineWeb2 benchmark performance
• Fully open: Emphasis on transparency

Dataset structure

Data fields

Each data entry includes the original FineWeb2 data fields with the addition of:

• quality_score: quality score obtained by the quality classifier
• embeddings: array of 768 floats containing mean-pooled XLM-RoBERTa embeddings for the first 512 tokens of the text

Data instance

{
  "date": "2022-11-27T02:08:37Z",
  "dump": "CC-MAIN-2022-49",
  "embeddings": [...],
  "file_path": "s3://commoncrawl/crawl-data/CC-MAIN-2022-49/segments/1669446710155.67/warc/CC-MAIN-20221127005113-20221127035113-00579.warc.gz",
  "id": "<urn:uuid:17cce4d8-e29c-4294-a5f1-315253b4e754>",
  "language": "fra",
  "language_score": 0.9986262321472168,
  "language_script": "Latn",
  "minhash_cluster_size": 1,
  "quality_score": 0.45224571228027344,
  "text": "Les grues du japon ,ou appelées localement tancho ,vivent sur l'ile\nd'Okkaido ,lors de la saison des amours\nElles effectuent dans la neige des parades nuptiales ,(digne d'un corps de ballet ),faites de sauts ,de battements d'ailes et de claquements de becs ,\nL'espéce , en voie d'extinction a la fin du vingtieme siecle ,a été sauvée par la population japonaise qui a nourri ces échassiers.\nLes grues ont longtemps fourni un sujet décoratif aux tapissiers nippons\nLes grues du Japon présentent un plumage entièrement blanc, si l'on excepte les rémiges secondaires et le cou, qui sont noirs. En Asie, ils ont un statut sacré car ils symbolisent l'amour.\nLes grues du japon ,ou appelées localement tancho ,vivent sur l'ile",
  "top_langs": {},
  "url": "http://http-jaouen-h-07.over-blog.fr/article-15193379.html",
}

Usage

You can load the dataset in Python using datasets:

from datasets import load_dataset

dataset = load_dataset("epfml/FineWeb2-HQ-Plus", "fra_Latn")

Licensing information

Like FineWeb2, this dataset is released under Open Data Commons Attribution License (ODC-By) v1.0 license and is subject to CommonCrawl's Terms of Use.

Dataset origin

Being a subset of FineWeb2, this data covers websites over the 2013-2024 time period.

FineWeb2 is sourced from the internet at large, it is very likely that some personable identifiable information (PII) will be present, even if the FineWeb2 processing has already anonymized email addresses and public IP addresses. If you find your own PII and would like it removed, please fill out the FineWeb2 PII removal/opt out form.

CommonCrawl respects robots.txt at crawl time, but if you are a webmaster and find your website in FineWeb2 and would like to have it removed, you may also use the FineWeb2 PII removal/opt out form.

Considerations for Using the Data

Before using this dataset for training models, we recommend performing additional filtering for sensitive content such as PII or harmful content. For the aspects of social impact, discussion of biases, and known limitations, we also refer to the FineWeb2 documentation.

Citation information

If you use this dataset in your research or applications, please use the following citation:

@misc{turki2026crosslingualqualityclassifiersmultilingual,
      title={Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection}, 
      author={Yassine Turki and Vinko Sabolčec and Bettina Messmer and Martin Jaggi},
      year={2026},
      eprint={2604.20549},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.20549}, 
}