The dataset contains 571 manually created instruction–response pairs in Faroese. The material covers a broad range of topics, with substantial content relating to the Faroe Islands, including Faroese language, history, society, geography, culture, literature, institutions, traditions and sports. It also contains questions and instructions relating to the Nordic countries, general knowledge, science, health, mathematics, law, taxation, travel and everyday topics.
The dataset contains both factual questions and more open-ended instructions, making it suitable for instruction tuning, supervised fine-tuning, question answering and evaluation of Faroese instruction-following language models.
The examples were independently created by three contributors. Each row includes a pseudonymous contributor identifier, allowing users to examine contributor-specific patterns and to identify cases where similar topics may have been covered independently by different contributors.
FO-Instruct is a Faroese instruction dataset in JSONL format developed at the Centre for Language Technology (Máltøknidepilin) at the University of the Faroe Islands (Fróðskaparsetur Føroya). It is available on Hugging Face.
Release: 2026
Authors: Iben Nyholm Debess, Uni Johannesen and Birita Dal-Christiansen
Developed at: Centre for Language Technology, University of the Faroe Islands
Contact: mtd@setur.fo
