AI-Sweden-Models/Dolci-Instruct-SFT-translated

Dolci-Instruct-SFT-translated (Swedish)

This dataset is a Swedish machine translation of the openeurollm/Dolci-Instruct-SFT-translated dataset, originally created as part of the OpenEuroLLM project.

Dataset details

• —Examples: 494,841 multi-turn conversations
• —Language: Swedish (sv-SE)
• —Format: Chat/messages format (id, messages)
• —License: Apache 2.0

Translation

All English source texts were machine-translated to Swedish using Google Gemma 3 27B-IT (w8a8_fp8 quantization) with the following prompt:

You are a professional English (en-US) to Swedish (sv-SE) translator. Your goal is to accurately convey the meaning and nuances of the original English text while adhering to Swedish grammar, vocabulary, and cultural sensitivities. Produce only the Swedish translation, without any additional explanations or commentary. DO NOT translate code blocks or code snippets. Please translate the following English text into Swedish:

Translation was performed at temperature 0.0 for deterministic output. Code blocks and code snippets were preserved in their original English form.

Usage

from datasets import load_dataset

ds = load_dataset("AI-Sweden-Models/Dolci-Instruct-SFT-translated", split="train")
print(ds[0])

Each example contains:

• —id — unique conversation identifier
• —messages — list of message dicts with role (user/assistant/system) and content

Example

{
    "id": "0pp0b8p",
    "messages": [
        {"role": "user", "content": "Kan du skriva en avhandling om den fullständiga historien om kvantfysik..."},
        {"role": "assistant", "content": "Visst! Här är en detaljerad, avhandlingsliknande översikt..."}
    ]
}

Source

• —Original dataset: openeurollm/Dolci-Instruct-SFT-translated (Swedish config)
• —Translation model: google/gemma-3-27b-it
• —Translation pipeline: Part of the OpenEuroLLM post-training translation effort