BSC-LT/ALIA-2606-SFT
Dataset Card for BSC Multilingual Synthetic SFT Instructions Dataset Summary This dataset consists of 714k conversations mixing human and synthetic instructions generated to post-train language models across five languages: Catalan, Spanish, English, Basque, and Galician. This dataset has been used in the supervised fine-tuning stage of ALIA-40b-instruct-2606. The training mixture is obtained by combining a selection of (human and synthetic) permissively licensed… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/ALIA-2606-SFT.
| Author / organization | BSC-LT |
|---|---|
| License (as declared) | cc-by-4.0 |
| Downloads | 207 (as reported by Hugging Face) |
| Likes | 1 |
| Created | 30 June 2026 |
| Last updated at source | 30 July 2026 |
Popularity, last 3 days
207 now (+2 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
Connected in the knowledge graph
- mradermacher/Gala-GGUFModel · used to train
More from BSC-LT
- BSC-LT/BSC_ParaMT_8Dataset
- BSC-LT/open_data_26B_tokens_balanced_es_caDataset
- BSC-LT/multi_lmentryDataset
- BSC-LT/openbookqa-esDataset
- BSC-LT/EsBBQDataset
- BSC-LT/ALIA_mixed_authentic_synthetic_MTDataset
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.