---
license: cc-by-nc-sa-4.0
task_categories:
- question-answering
language:
- ko
- en
tags:
- synthetic
- sft
- dpo
- instruction following
size_categories:
- 100K<n<1M
configs:
- config_name: korean
  data_files:
  - split: train
    path: korean/train.json
- config_name: english
  data_files:
  - split: train
    path: english/train.json
---

# Dataset Card for SNU Thunder Synthetic InstructionFollowing

## Dataset Summary

This dataset was used as part of the post-training corpus for SnuLLM(`to_fill`).

This dataset consists of Korean and English question-answer pairs. Questions are sourced from publicly available datasets, and answers were generated using open large language models (Exaone 3.5, LLaMA 3.3, Qwen 2.5). It is intended for research and non-commercial use.

## Supported Tasks

**Tasks**: Instruction following

## Languages

- Korean (ko)
- English (en)

---

## Dataset Structure

### Dataset Instance Example

- Korean
  ```json
  {
    "question_source":"KoAlpaca",
    "question":"씨앗을 냉장 처리후 심는 이유와 장기보존하려면 어떻게 해야 할까요?\n답변은 영어로 작성해야 하며 다른 언어는 허용되지 않습니다.",
    "response":"Seeds are often refrigerated before planting, a process known as stratification, to simulate the natural winter period that some seeds require to germinate. This cold treatment can help break seed dormancy, increase germination rates, and promote healthy seedling growth.\n\nTo stratify seeds, you can store them in a sealed container or plastic bag in the refrigerator at a temperature between 32°F (0°C) and 40°F (4°C) for several weeks or months, depending on the seed type. Some seeds may require a specific duration of cold stratification, so it's essential to research the particular needs of your seeds.\n\nFor long-term preservation, seeds can be stored in a cool, dry place, such as a basement or a room with a consistent temperature between 40°F (4°C) and 50°F (10°C). You can store seeds in:\n\n1. Airtight containers: Glass jars or plastic containers with tight-fitting lids can help maintain a dry environment.\n2. Paper envelopes: Seeds can be stored in paper envelopes, which allow for air circulation and can help prevent moisture buildup.\n3. Seed vaults: Specifically designed seed vaults or seed banks can provide optimal conditions for long-term seed storage, with controlled temperature, humidity, and light.\n\nTo maintain seed viability, it's crucial to:\n\n1. Keep seeds dry: Moisture can lead to seed degradation, so it's essential to store seeds in a dry environment.\n2. Maintain a consistent temperature: Avoid exposing seeds to extreme temperatures, which can affect their viability.\n3. Protect from light: Light can cause seeds to degrade, so it's best to store them in a dark or low-light environment.\n4. Use desiccants: Desiccants, such as silica gel, can help maintain a dry environment and prevent moisture buildup.\n\nBy following these guidelines, you can help preserve seeds for extended periods and ensure their viability for future planting.",
    "model":"Llama-3.3-70B-Instruct",
    "chosen":true
  },
  ```
- English
  ```json
  {
    "question_source":"SlimOrca",
    "question":"If \"5 guys in front of a black volkswagon car and 2 guys are holding white boxes.\" does that mean that \"The vw is a ford.\"?\nOptions:\n- yes\n- it is not possible to tell\n- no A step-by-step solution is:\n\nAt the end of your response, please explicitly add a postscript starting with P.S.",
    "response":"Step 1: Analyze the given statement: \"5 guys in front of a black volkswagon car and 2 guys are holding white boxes.\"\nStep 2: Identify the key information: There is a black Volkswagen car.\nStep 3: Compare this information to the question: \"Does that mean that the vw is a ford?\"\nStep 4: Determine if the information provided about the car (being a Volkswagen) supports or contradicts the claim that it is a Ford.\nStep 5: Conclude that since the car is explicitly described as a Volkswagen, it cannot be a Ford as these are different car brands.\n\nThe correct answer is: no\n\nP.S. The statement clearly identifies the car as a Volkswagen, which directly contradicts the notion that it could be a Ford.",
    "model":"Qwen2.5-32B-Instruct",
    "chosen":true
  },
  ```

### Data Fields
- `question_source`: Source dataset of the question.
- `question`: The question.
- `response`: The answer generated by the `model`.
- `model`: The model used to generate the answer.
- `chosen`: A boolean field (`true` or `false`) indicating whether the corresponding answer was selected as the preferred response.
  - `true`: The answer was used as the chosen (accepted) response.
  - `false`: The answer was used as a rejected alternative (e.g., in preference tuning).

### Data Splits

Each subset (`english`, `korean`) only includes a `train` split.

| Subset  | Split  | Size   |
|---------|--------|--------|
| korean  | train  | 59948  |
| english | train  | 361594 |

---

## Dataset Creation

### Source Data

Questions were sourced from the following datasets:
- [KoAlpaca](https://github.com/Beomi/KoAlpaca)
- [SlimOrca](https://huggingface.co/datasets/Open-Orca/SlimOrca)

### Generation

Answers were generated using the following open LLMs:
- [EXAONE-3.5-32B-Instruct](https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct)
- [Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct)
- [Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct)

Each question was passed to a model prompt, and one or more generated responses were collected.

For more details on the generation configuration, please refer to our paper:
`to_fill`

### Annotation

A subset of responses was classified as chosen/rejected responses by rule-based evaluation using the [`lm-eval-harness`](https://github.com/EleutherAI/lm-eval-harness) framework.

---

## Additional Information

### Licensing Information

This dataset is released under the CC-BY-NC-SA-4.0 license.

### Citation Information

#### Our Citation Information

`to_fill`

#### Original Datasets Citation Information

- KoAlpaca
  ```bibtex
  @misc{alpaca,
    author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
    title = {Stanford Alpaca: An Instruction-following LLaMA model},
    year = {2023},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
  }
  ```
- SlimOrca
  ```bibtex
  @misc{SlimOrca,
    title = {SlimOrca: An Open Dataset of GPT-4 Augmented FLAN Reasoning Traces, with Verification},
    author = {Wing Lian and Guan Wang and Bleys Goodson and Eugene Pentland and Austin Cook and Chanvichet Vong and "Teknium"},
    year = {2023},
    publisher = {HuggingFace},
    url = {https://https://huggingface.co/Open-Orca/SlimOrca}
  }
  @misc{mukherjee2023orca,
        title={Orca: Progressive Learning from Complex Explanation Traces of GPT-4}, 
        author={Subhabrata Mukherjee and Arindam Mitra and Ganesh Jawahar and Sahaj Agarwal and Hamid Palangi and Ahmed Awadallah},
        year={2023},
        eprint={2306.02707},
        archivePrefix={arXiv},
        primaryClass={cs.CL}
  }
  @misc{longpre2023flan,
        title={The Flan Collection: Designing Data and Methods for Effective Instruction Tuning}, 
        author={Shayne Longpre and Le Hou and Tu Vu and Albert Webson and Hyung Won Chung and Yi Tay and Denny Zhou and Quoc V. Le and Barret Zoph and Jason Wei and Adam Roberts},
        year={2023},
        eprint={2301.13688},
        archivePrefix={arXiv},
        primaryClass={cs.AI}
  }
  ```

#### Generation Models Citation Information

- EXAONE-3.5-32B-Instruct
  ```bibtex
  @article{exaone-3.5,
    title={EXAONE 3.5: Series of Large Language Models for Real-world Use Cases},
    author={LG AI Research},
    journal={arXiv preprint arXiv:https://arxiv.org/abs/2412.04862},
    year={2024}
  }
  ```
- Llama-3.3-70B-Instruct
  ```bibtex
  @article{grattafiori2024llama,
    title={The llama 3 herd of models},
    author={Grattafiori, Aaron and Dubey, Abhimanyu and Jauhri, Abhinav and Pandey, Abhinav and Kadian, Abhishek and Al-Dahle, Ahmad and Letman, Aiesha and Mathur, Akhil and Schelten, Alan and Vaughan, Alex and others},
    journal={arXiv preprint arXiv:2407.21783},
    year={2024}
  }
  ```
- Qwen2.5-32B-Instruct
  ```bibtex
  @misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
  }
  @article{qwen2,
    title={Qwen2 Technical Report}, 
    author={An Yang and Baosong Yang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Zhou and Chengpeng Li and Chengyuan Li and Dayiheng Liu and Fei Huang and Guanting Dong and Haoran Wei and Huan Lin and Jialong Tang and Jialin Wang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Ma and Jin Xu and Jingren Zhou and Jinze Bai and Jinzheng He and Junyang Lin and Kai Dang and Keming Lu and Keqin Chen and Kexin Yang and Mei Li and Mingfeng Xue and Na Ni and Pei Zhang and Peng Wang and Ru Peng and Rui Men and Ruize Gao and Runji Lin and Shijie Wang and Shuai Bai and Sinan Tan and Tianhang Zhu and Tianhao Li and Tianyu Liu and Wenbin Ge and Xiaodong Deng and Xiaohuan Zhou and Xingzhang Ren and Xinyu Zhang and Xipin Wei and Xuancheng Ren and Yang Fan and Yang Yao and Yichang Zhang and Yu Wan and Yunfei Chu and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zhihao Fan},
    journal={arXiv preprint arXiv:2407.10671},
    year={2024}
  }
  ```

### Contact Information

If you find something wrong or have question about the dataset, contact snullm@aces.snu.ac.kr.