V4ldeLund/scandi-translated-instruct

---
pretty_name: scandi-translated-instruct
language:
- da
- sv
- nn
- nb
tags:
- instruction-tuning
- chat
- conversational
- machine-translation
- nordic
- multilingual
task_categories:
- text-generation
size_categories:
- 1M<n<10M
license: other
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: language
dtype: string
- name: model
dtype: string
- name: messages
list:
- name: content
dtype: string
- name: role
dtype: string
- name: source
dtype: string
---

# scandi-translate

A Scandinavian instruction‑tuning dataset built from machine‑translated instruction/response pairs. It unifies Danish, Swedish, Norwegian Bokmål and Norwegian Nynorsk data into a chat-friendly `messages` schema.

## Dataset size

- Total rows: **1,252,683**
- Per-language counts measured during build:
- Danish (`da`): **377,413**
- Swedish (`sv`): **332,558**
- Norwegian Nynorsk (`nn`): **270,772**
- Norwegian Bokmål (`nb`): **268,875**
- By source:
- `V4ldeLund/da-translated-instruct`
- `akoksal/muri-it-language-split` (configs `swe`, `nor`)
- `CohereLabs/aya_collection_language_split` (configs `swedish`, `norwegian_bokmal`, `norwegian_nynorsk`)
- `neph1/Alpaca-Lora-GPT4-Swedish-Refined` (train split)

## Licences

| Source dataset | License note (as of 2026‑01‑10) |
| --- | --- |
| `akoksal/muri-it-language-split` (swe/nor) | License **Apache-2.0**
| `CohereLabs/aya_collection_language_split` | **Apache-2.0**.
| `neph1/Alpaca-Lora-GPT4-Swedish-Refined` | License not listed on HF page
| `V4ldeLund/da-translated-instruct` (contains `akoksal/muri-it-language-split` dan, `Mabeck/danish-OpenHermes`, `CohereLabs/aya_collection_language_split` danish) | Mixed Apache-2.0 / MIT

Because of the mixture and unspecified items, the combined release is marked **“other”**. Every row keeps its `source` so downstream users can honor upstream terms.

## Languages included

- Danish (`da`)
- Swedish (`sv`)
- Norwegian Nynorsk (`nn`)
- Norwegian Bokmål (`nb`)

## Format

Columns

- `language` (string): ISO two letter language code (`da`, `sv`, `nn`, `nb`).
- `model` (string): translation/model identifier used when generating the pair.
- `source` (string): upstream dataset repo id.
- `messages` (list): chat messages with
- `role`: one of `system`, `user`, `assistant`
- `content`: message text

Example

```json
{
"model": "Mixtral-8x7B",
"source": "akoksal/muri-it-language-split",
"language": "nb",
"messages": [
{"role": "user", "content": "Forklar kort fotosyntesen."},
{"role": "assistant", "content": "Fotosyntese er processen..." }
]
}
```