HiTZ Center, University of the Basque Country
lab · SpainScores
1 product on the map — 1 open.
Openness
4 high confidence- license
- mixed-per-subset (No dataset-level license on either card or in the Hugging Face metadata; each document carries its source license in its meta field. The MIT LICENSE in the GitHub repository covers the training code.)
- access
- public (Hugging Face gated: false on both repositories.)
- dataset_card
- present (Sources, per-source document counts and funding on both cards.)
Both repositories download without a gate, but the corpus has no license of its own: each document keeps the terms of the crawl or public dataset it came from. The maintainers removed documents they lacked permission to redistribute, so the public copy is a curated subset of what the models saw.
- https://huggingface.co/api/datasets/HiTZ/latxa-corpus-v1.1 recorded 2026-09-24
"gated":false, no license in cardData
- https://huggingface.co/api/datasets/HiTZ/latxa-corpus-v2 recorded 2026-09-24
"gated":false, no license in cardData
- https://huggingface.co/datasets/HiTZ/latxa-corpus-v1.1/raw/main/README.md recorded 2026-09-24
v1.1 card lists five sources and says to consult each corpus's references for licenses; same permission-compliance notice.
- https://huggingface.co/datasets/HiTZ/latxa-corpus-v2/raw/main/README.md recorded 2026-09-24
"For detailed information regarding the licenses associated with each individual corpus and document ... please refer to the 'license' field within 'meta'"; notice that "Some data points have been removed to ensure permission compliance."
- https://raw.githubusercontent.com/hitz-zentroa/latxa/main/LICENSE recorded 2026-09-24
MIT License, Copyright (c) 2024 HiTZ zentroa; the README describes the repository as training code.
Adoption
1 high confidenceHugging Face downloads over the trailing 30 days, summed across the v1.1 and v2 repositories. A download is a file fetch rather than a trained model, and copies mirrored elsewhere are not counted.
- https://huggingface.co/api/datasets/HiTZ/latxa-corpus-v1.1 recorded 2026-09-24
565 downloads in the trailing 30 days for HiTZ/latxa-corpus-v1.1
- https://huggingface.co/api/datasets/HiTZ/latxa-corpus-v2 recorded 2026-09-24
328 downloads in the trailing 30 days for HiTZ/latxa-corpus-v2
Capability
3 medium confidenceLatxa Corpus is the pretraining set behind every Latxa model for Basque, joining a new Basque web crawl and institutional text with the Basque parts of large web corpora, documented in an ACL paper and two cards. Its four billion tokens are judged against what Basque has, an official language with a modest written web whose speaker count is uncertain around a million; the stricter reading of that uncertainty is used, and the more generous one would place it one step higher.
- https://arxiv.org/abs/2403.20266 recorded 2026-11-11
"we continue pretraining on a new Basque corpus comprising 4.3M documents and 4.2B tokens"
- https://huggingface.co/api/models?filter=dataset:HiTZ/latxa-corpus-v1.1 recorded 2026-11-11
Models tagged with the dataset include HiTZ/latxa-7b-v1.1, latxa-13b-v1.1, latxa-70b-v1.1 and the v1.2 releases.
- https://huggingface.co/api/models?filter=dataset:HiTZ/latxa-corpus-v2 recorded 2026-11-11
Models tagged with the dataset include HiTZ/Latxa-Llama-3.1-70B-Instruct-v2, HiTZ/Latxa-Qwen3.5-4B and HiTZ/Latxa-Qwen3.5-2B.
- https://huggingface.co/datasets/HiTZ/latxa-corpus-v1.1/raw/main/README.md recorded 2026-11-11
"This is the training corpus of Latxa v1.1, a family of large language models for Basque based on Llama 2."
- https://huggingface.co/datasets/HiTZ/latxa-corpus-v2/raw/main/README.md recorded 2026-11-11
Data Statistics table: per-source train/valid/test document counts for 14 sources, EusCrawl v2 largest at 1,636,010 train documents.