ctauchmann/dolci-4-lang

Dolci 4-Language (Machine-Translated)

Dataset Summary

This dataset is a 4-language machine-translated derivative of the AllenAI Dolci dataset collection, used in the OLMo 3 post-training pipeline. It contains translations of reasoning, instruction-following, and alignment data into German, French, Italian, and Spanish for research on multilingual reasoning and alignment.

Translations were generated automatically using Qwen/Qwen3-Next-80B-A3B-Instruct and may contain errors.

Dataset Format

All files use a standardized 5-column JSONL format:

Column	Description
source_id	Join key to English canonical dataset
output_id	Unique identifier per translation
language	Target language (de/es/fr/it)
dataset_source	Upstream HuggingFace dataset name
message	Translated text

Translation Completion Status

Dolci-Think-SFT-7B-4-Lang (source: 2,268,178 samples)

Language	Samples	% Complete
DE	2,221,369	98%
ES	1,560,000	69%
FR	1,272,000	56%
IT	1,510,000	67%

Dolci-Think-DPO-7B-4-Lang (source: 150,000 samples)

Language	Samples	% Complete
DE	150,067	100%
ES	149,997	100%
FR	149,997	100%
IT	149,997	100%

Dolci-Think-RL-7B-4-Lang (source: 102,014 samples)

Language	Samples	% Complete
DE	79,336	78%
ES	79,337	78%
FR	79,334	78%
IT	79,323	78%

Dolci-Think-RL-7B-Completions-SFT-4-Lang (source: 636,095 samples)

Language	Samples	% Complete
DE	288,822	45%
ES	293,455	46%
FR	291,159	46%
IT	291,870	46%

Dolci-Think-RL-7B-Completions-DPO-4-Lang (source: 556,095 samples)

Language	Samples	% Complete
DE	165,025	30%
ES	165,024	30%
FR	165,202	30%
IT	165,184	30%

Dolci-Instruct-SFT-4-Lang (source: 2,152,112 samples)

Language	Samples	% Complete
DE	556,066	26%
ES	558,281	26%
FR	557,759	26%
IT	556,733	26%

Dataset Contents

Subset	Original Source	EN Samples	Type
Dolci-Think-SFT-7B-4-Lang	Dolci-Think-SFT-7B	2,268,178	SFT
Dolci-Think-DPO-7B-4-Lang	Dolci-Think-DPO-7B	150,000	DPO
Dolci-Think-RL-7B-Completions-SFT-4-Lang	Dolci-Think-RL-7B-Completions-SFT	636,095	SFT
Dolci-Think-RL-7B-Completions-DPO-4-Lang	Dolci-Think-RL-7B-Completions-DPO	556,095	DPO
Dolci-Think-RL-7B-4-Lang	Dolci-Think-RL-7B	102,014	RL
Dolci-Instruct-SFT-4-Lang	Dolci-Instruct-SFT	2,152,112	SFT

Languages

Each subset contains translations in 4 languages:

• de.jsonl - German (Deutsch)
• es.jsonl - Spanish (Español)
• fr.jsonl - French (Français)
• it.jsonl - Italian (Italiano)

Translation Details

• Model: Qwen/Qwen3-Next-80B-A3B-Instruct
• Method: Parallel inference via vLLM fleet (H100 GPUs)
• Date: January-February 2026
• Post-editing: None (fully automatic)

The translation preserves:

• Mathematical notation and LaTeX
• Code blocks and syntax highlighting
• XML tags (<think>, <answer>, etc.)
• Logical structure of reasoning chains

Usage

from datasets import load_dataset

# Load a specific subset and language
dataset = load_dataset(
    "AIML-TUDA/dolci-4-lang",
    data_files="Dolci-Think-SFT-7B-4-Lang/de.jsonl"
)

# Load all German data
dataset = load_dataset(
    "AIML-TUDA/dolci-4-lang",
    data_files="*/de.jsonl"
)

# Download via CLI
huggingface-cli download AIML-TUDA/dolci-4-lang \
    Dolci-Think-SFT-7B-4-Lang/de.jsonl \
    --local-dir ./data \
    --repo-type dataset

Intended Use

This dataset is intended for research and educational use, in accordance with Ai2's Responsible Use Guidelines. We kindly request that it not be used for commercial, surveillance, or harmful applications.

Provenance

Source: AllenAI Dolci dataset collection (ODC-BY)

• allenai/Dolci-Think-SFT-7B
• allenai/Dolci-Think-DPO-7B
• allenai/Dolci-Think-RL-7B
• allenai/Dolci-Think-RL-7B-Completions-SFT
• allenai/Dolci-Think-RL-7B-Completions-DPO
• allenai/Dolci-Instruct-SFT

Translation model: Qwen/Qwen3-Next-80B-A3B-Instruct (Apache-2.0)

Translations were generated automatically without human post-editing.

License

This dataset is a derivative of the AllenAI Dolci dataset collection and is released under the same license: ODC-BY.

Original license statement (AllenAI):

This dataset is licensed under ODC-BY. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines.

Citation

If you use this dataset, please cite both the original AllenAI dataset and this derived dataset.

Original Dataset (OLMo 3)

@misc{olmo2025olmo3,
  title        = {OLMo 3},
  author       = {Team OLMo and Allyson Ettinger and Amanda Bertsch and Bailey Kuehl and David Graham and David Heineman and Dirk Groeneveld and Faeze Brahman and Finbarr Timbers and Hamish Ivison and Jacob Morrison and Jake Poznanski and Kyle Lo and Luca Soldaini and Matt Jordan and Mayee Chen and Michael Noukhovitch and Nathan Lambert and Pete Walsh and Pradeep Dasigi and Robert Berry and Saumya Malik and Saurabh Shah and Scott Geng and Shane Arora and Shashank Gupta and Taira Anderson and Teng Xiao and Tyler Murray and Tyler Romero and Victoria Graf and Akari Asai and Akshita Bhagia and Alexander Wettig and Alisa Liu and Aman Rangapur and Chloe Anastasiades and Costa Huang and Dustin Schwenk and Harsh Trivedi and Ian Magnusson and Jaron Lochner and Jiacheng Liu and Lester James V. Miranda and Maarten Sap and Malia Morgan and Michael Schmitz and Michal Guerquin and Michael Wilson and Regan Huff and Ronan Le Bras and Rui Xin and Rulin Shao and Sam Skjonsberg and Shannon Zejiang Shen and Shuyue Stella Li and Tucker Wilde and Valentina Pyatkin and Will Merrill and Yapei Chang and Yuling Gu and Zhiyuan Zeng and Ashish Sabharwal and Luke Zettlemoyer and Pang Wei Koh and Ali Farhadi and Noah A. Smith and Hannaneh Hajishirzi},
  year         = {2025},
  eprint       = {2512.13961},
  archivePrefix= {arXiv},
  primaryClass = {cs.CL},
  url          = {https://arxiv.org/abs/2512.13961}
}

This Dataset

@misc{dolci_4lang_2025,
  title        = {Dolci 4-Language: Machine-Translated Reasoning Data},
  author       = {Christopher Tauchmann},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/AIML-TUDA/dolci-4-lang}},
  note         = {Derived from AllenAI Dolci (ODC-BY); translations generated with Qwen/Qwen3-Next-80B-A3B-Instruct}
}