Datasets:

HiTZ
/

latxa-corpus-v2

language:
  - eu
configs:
  - config_name: wikipedia
    data_files:
      - split: train
        path: wikipedia/train*.jsonl.gz
      - split: validation
        path: wikipedia/valid*.jsonl.gz
      - split: test
        path: wikipedia/test*.jsonl.gz
  - config_name: euscrawl-v2
    data_files:
      - split: train
        path: euscrawl-v2-free/train*.jsonl.gz
      - split: validation
        path: euscrawl-v2-free/valid*.jsonl.gz
      - split: test
        path: euscrawl-v2-free/test*.jsonl.gz
  - config_name: zelaihandi
    data_files:
      - split: train
        path: zelaihandi/train*.jsonl.gz
      - split: validation
        path: zelaihandi/valid*.jsonl.gz
      - split: test
        path: zelaihandi/test*.jsonl.gz
  - config_name: bog
    data_files:
      - split: train
        path: bog/train*.jsonl.gz
      - split: validation
        path: bog/valid*.jsonl.gz
      - split: test
        path: bog/test*.jsonl.gz
  - config_name: bopv
    data_files:
      - split: train
        path: bopv/train*.jsonl.gz
      - split: validation
        path: bopv/valid*.jsonl.gz
      - split: test
        path: bopv/test*.jsonl.gz
  - config_name: botha
    data_files:
      - split: train
        path: botha/train*.jsonl.gz
      - split: validation
        path: botha/valid*.jsonl.gz
      - split: test
        path: botha/test*.jsonl.gz
  - config_name: parleus
    data_files:
      - split: train
        path: parleus/train*.jsonl.gz
      - split: validation
        path: parleus/valid*.jsonl.gz
      - split: test
        path: parleus/test*.jsonl.gz
  - config_name: aldizkariak
    data_files:
      - split: train
        path: aldizkariak/train*.jsonl.gz
      - split: validation
        path: aldizkariak/valid*.jsonl.gz
      - split: test
        path: aldizkariak/test*.jsonl.gz
  - config_name: cultura-x
    data_files:
      - split: train
        path: cultura-x/train*.jsonl.gz
      - split: validation
        path: cultura-x/valid*.jsonl.gz
      - split: test
        path: cultura-x/test*.jsonl.gz
  - config_name: hplt-v2
    data_files:
      - split: train
        path: hplt-v2/train*.jsonl.gz
      - split: validation
        path: hplt-v2/valid*.jsonl.gz
      - split: test
        path: hplt-v2/test*.jsonl.gz
  - config_name: finepdfs
    data_files:
      - split: train
        path: finepdfs/train*.jsonl.gz
      - split: validation
        path: finepdfs/valid*.jsonl.gz
      - split: test
        path: finepdfs/test*.jsonl.gz
  - config_name: fineweb2
    data_files:
      - split: train
        path: fineweb2/train*.jsonl.gz
      - split: validation
        path: fineweb2/valid*.jsonl.gz
      - split: test
        path: fineweb2/test*.jsonl.gz
  - config_name: colossal-oscar
    data_files:
      - split: train
        path: colossal-oscar/train*.jsonl.gz
      - split: validation
        path: colossal-oscar/valid*.jsonl.gz
      - split: test
        path: colossal-oscar/test*.jsonl.gz
  - config_name: hplt-v1
    data_files:
      - split: train
        path: hplt-v1/train*.jsonl.gz
      - split: validation
        path: hplt-v1/valid*.jsonl.gz
      - split: test
        path: hplt-v1/test*.jsonl.gz
task_categories:
  - fill-mask
  - text-generation
task_ids:
  - language-modeling
  - masked-language-modeling
annotations_creators:
  - no-annotation
multilinguality:
  - monolingual

Latxa Corpus v2

📧 Point of Contact: hitz@ehu.eus

Dataset Summary

Curated by: HiTZ Research Center & IXA Research group (University of the Basque Country UPV/EHU)
Language(s): eu-ES

Latxa Corpus v2 is a large-scale monolingual Basque corpus, created by combining curated crawls, public datasets, institutional data, and newly collected resources. Compared to v1.1, it substantially increases coverage, diversity, and volume. The final corpus is deduplicated, filtered, and ready for language model pretraining.

📌 Notice
As of February 13th 2026, this repository reflects a curated version of the original dataset. Some data points have been removed to ensure permission compliance. Researchers with a legitimate need for access to the original data may contact the maintainers for further information.

Data Sources

Euscrawl v2: An updated version of EusCrawl v1 (Artetxe et al., 2022), including new content up to November 2025.
ZelaiHandi: A subset of the ZelaiHandi corpus.
BOPV: The official gazette of the Basque Government (https://opendata.euskadi.eus/api-bopv/?api=bopv).
BOG: The official gazette of the Provincial Council of Gipuzkoa (https://egoitza.gipuzkoa.eus/es/bog).
BOTHA: The official gazette of the Provincial Council of Álava (https://www.araba.eus/botha).
ParlEus: Transcriptions of Basque Parliament sessions, extracted and reconstructed from PDFs (https://www.legebiltzarra.eus).
Aldizkariak: Basque academic journals processed from PDF sources:
- Ekaia: https://ojs.ehu.eus/index.php/ekaia
- Gogoa: https://ojs.ehu.eus/index.php/Gogoa
- Tantak: https://ojs.ehu.eus/index.php/Tantak
- Ekonomiaz: https://www.euskadi.eus/web01-a2reveko/es/k86aEkonomiazWar/ekonomiaz
- Uztaro: https://aldizkariak.ueu.eus/index.php/uztaro
- IkerGazte: https://www.ueu.eus/ikergazte/2025/artikuluak/argitalpenak
Wikipedia: Basque Wikipedia's dump from September 2025.
CulturaX: The Basque portion of the CulturaX corpus (Nguyen et al., 2023).
Colossal OSCAR: The Basque portion of several Colossal OSCAR releases.
FineWeb2: The Basque portion of the FineWeb2 corpus (Penedo et al., 2024).
FinePDFs: The Basque portion of the FinePDFs corpus.
HPLT v1 & v2: The Basque portion of the HPLT v1 (Aulamo et al., 2023) and HPLT v2 corpora.

Data Statistics

The size of each dataset in terms of number of documents can be found below:

	Train	Valid	Test
Aldizkariak	3,260	33	33
BOG	169,665	1,731	1,731
BOPV	42,798	437	887
BOTHA	92,249	941	941
Colossal OSCAR	89,856	915	915
CulturaX	870,464	8,882	8,882
Euscrawl v2	1,636,010	16,682	17,097
FinePDFs	277,572	2,803	837
FineWeb2	341,296	3,482	3,482
HPLT v1	894,590	9,128	9,128
HPLT v2	567,503	5,790	5,790
ParlEus	22,818	231	560
Wikipedia	456,273	4,655	4,655
ZelaiHandi	189,975	1,938	1,938

Licensing

We do not claim ownership of any document in the corpus. Should you consider that our data contains material that is owned by you and you would not like to be reproduced here, please contact us at hitz@ehu.eus. For detailed information regarding the licenses associated with each individual corpus and document comprising this training dataset, please refer to the "license" field within "meta" or the respective references listed alongside each corpus entry.

Funding

This work is funded by the Basque Government (IKER-GAITU project) and the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ILENIA with reference 2022/TL22/00215335 and within the framework of the project Desarrollo de Modelos ALIA.