File size: 11,066 Bytes
c1a87f8 d73fe8d c1a87f8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 | ---
license: other
pretty_name: nvidia/ProfBench (mirror)
source_datasets:
- nvidia/ProfBench
tags:
- laya
- laya-finetune
- mirror
- domain-science
- domain-business
- use-conditional
- genre-exam
- label-text
- annot-unknown
- lang-en
language:
- en
size_categories:
- n<1K
license_name: nvidia-evaluation-dataset-license
---
# nvidia/ProfBench(複製)
## データの内容
人手でアノテーションされた 40 件の専門的な課題と、それに対する 3 モデルの応答、評価基準のデータ。1 行は 1 課題で、`prompt` に課題文、`rubrics` に 15〜59 個の評価基準、`domain` に領域、`o3_response` などに各モデルの応答が入る。領域は Chemistry PhD、Physics PhD、Finance MBA、Consulting MBA の 4 種類で、一部の応答は Grok を使って作られており、カードは LLM の評価に使うことを想定している。
## データの概要
| 項目 | 値 |
|---|---|
| 行数 | 40(1 構成・1 分割の合計) |
| 列数 | 7 |
| 容量 | 1.20 MB(ファイル 4 本) |
| ファイルの形式 | pdf 1、jsonl 1 |
| 言語 | en |
| 公開 | public |
### 構成と分割
| 構成(config) | 分割(split)ごとの行数 | 列数 |
|---|---|---:|
| `default` | `test` 40 | 7 |
### 列(構成 `default`・分割 `test`)
| 列 | 型 | 先頭の行の値 |
|---|---|---|
| `task_id` | `string` | `Chem-0` |
| `domain` | `string` | `Chemistry PhD` |
| `prompt` | `string` | `An experimental procedure using UV-Vis for the d…` |
| `o3_response` | `string` | `Step-1: Hydrazine concentration from the Watt–Ch…` |
| `r1-0528_response` | `string` | `To calculate the **Faradaic efficiency (FE)** an…` |
| `grok4_response` | `string` | `The Faradaic efficiency is calculated using the …` |
| `rubrics` | `list<struct{criterion_description: string, criterion_weight: string, criterion_type: list<string>, o3_fulfilment: bool, r1-0528_fulfilment: bool, …}>` | `[23 件]` |
行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は [HYUKI5/dataset-catalog](https://huggingface.co/datasets/HYUKI5/dataset-catalog) にある。
## 複製について
[`nvidia/ProfBench`](https://huggingface.co/datasets/nvidia/ProfBench) をリビジョン `f5c471cf2f277c264621e4b5c0cad5bb466f9260` の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(`convaiinnovations/laya`)の微調整に使う候補として、調査メモ(`Laya 学習用データセット調査`)に載せたデータを集めている。
## 出典
- 元のデータセット: https://huggingface.co/datasets/nvidia/ProfBench
- 複製したリビジョン: `f5c471cf2f277c264621e4b5c0cad5bb466f9260`(2026-03-04 06:27:42+00:00)
- 元のライセンス表記: `other`
## 論文
- https://arxiv.org/abs/2510.18941
## アノテーションの方法
**データセットの作成日:**
9/24/2025
## 微調整に使うときの注意
- 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
## 調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
| `nvidia_nemotron.md` | Q2. 金融・経済・ビジネス・ニュース・法務に関係するサブセットはあるか。ラベルはあるか (L79) | nvidia-evaluation-dataset-license(評価用。学習への利用可否は条文未読で未確認) | 確認済 | 原文注記: 40 件のみ。評価専用の可能性が高く、学習に入れるのは避ける |
## 元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
---
## データセットの説明:
[Leaderboard](https://huggingface.co/spaces/nvidia/ProfBench) | [ブログ](https://huggingface.co/blog/nvidia/profbench) | [論文](https://arxiv.org/abs/2510.18941) | [データ](https://huggingface.co/datasets/nvidia/ProfBench) | [コード](https://github.com/NVlabs/ProfBench) | [Nemo Evaluator SDK](https://github.com/NVIDIA-NeMo/Evaluator)
[](https://www.youtube.com/watch?v=GEPvdq3C54s)
40 件の人手でアノテーションされた課題にまたがる、3000 組を超えるルーブリックと応答の組。専門的な課題に答えたレポートを扱い、対象領域は博士課程レベルの STEM(化学、物理)と専門サービス(金融サービス、経営コンサルティング)である。
このデータセットは商用・非商用のどちらにも使える。
## データセットの所有者:
NVIDIA Corporation
## データセットの作成日:
9/24/2025
## ライセンス/利用規約:
NVIDIA Evaluation Dataset License
## 想定される用途:
専門的な課題で LLM を評価したい研究者と開発者。ProfBench は、数十のベンチマークを横断する評価のための統一インターフェースを備えた [Nemo Evaluator SDK](https://github.com/NVIDIA-NeMo/Evaluator) の一部として使うことを推奨する。
## データセットの特徴づけ:
** データ収集の方法<br>
* [ハイブリッド: 人手・合成・自動]<br>
** ラベル付けの方法<br>
* [人手] <br>
## データセットの形式:
テキスト。
## データセットの量:
40 件のレコード
各レコードには次のフィールドが含まれる:
- ID: 各サンプルの一意の識別子
- Domain: 領域。Chemistry PhD / Physics PhD / Finance MBA / Consulting MBA のいずれか
- Prompt: 大規模言語モデル(LLM)への指示
- Rubrics: 最終的なモデル出力を評価するために使う、15〜59 個の一意の基準
- Model Responses: OpenAI o3 / xAI Grok4 / DeepSeek R1-0528 による 3 件の応答
このデータセットの一部は Grok を使って作成された。
総容量: 1 MB。
## 倫理的な配慮:
NVIDIA は、信頼できる AI は共通の責任であると考えており、広範な AI アプリケーションの開発を可能にするための方針と慣行を整えている。利用規約に従ってダウンロードまたは使用する場合、開発者は社内のモデルチームと協力して、このモデルが関連する業界と用途の要件を満たし、想定外の製品の誤用に対処することを確認すべきである。
モデルの品質、リスク、セキュリティ上の脆弱性、または NVIDIA の AI に関する懸念は、[こちら](https://www.nvidia.com/en-us/support/submit-security-vulnerability/) から報告すること。
## 引用:
ProfBench が役に立った場合は、以下を引用することを検討すること:
```
@misc{wang2025profbenchmultidomainrubricsrequiring,
title={ProfBench: Multi-Domain Rubrics requiring Professional Knowledge to Answer and Judge},
author={Zhilin Wang and Jaehun Jung and Ximing Lu and Shizhe Diao and Ellie Evans and Jiaqi Zeng and Pavlo Molchanov and Yejin Choi and Jan Kautz and Yi Dong},
year={2025},
eprint={2510.18941},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.18941},
}
```
<details>
<summary>原文(英語)</summary>
## Dataset Description:
[Leaderboard](https://huggingface.co/spaces/nvidia/ProfBench) | [Blog](https://huggingface.co/blog/nvidia/profbench) | [Paper](https://arxiv.org/abs/2510.18941) | [Data](https://huggingface.co/datasets/nvidia/ProfBench) | [Code](https://github.com/NVlabs/ProfBench) | [Nemo Evaluator SDK](https://github.com/NVIDIA-NeMo/Evaluator)
[](https://www.youtube.com/watch?v=GEPvdq3C54s)
More than 3000 rubric-response pairs across 40 human-annotated tasks presenting reports addressing professional tasks across PhD STEM (Chemistry, Physics) and Professional Services (Financial Services, Management Consulting) domains.
This dataset is ready for commercial/non-commercial use.
## Dataset Owner(s):
NVIDIA Corporation
## Dataset Creation Date:
9/24/2025
## License/Terms of Use:
NVIDIA Evaluation Dataset License
## Intended Usage:
Researchers and developers seeking to evaluate LLMs on Professional Tasks. We recommend use of ProfBench as part of [Nemo Evaluator SDK](https://github.com/NVIDIA-NeMo/Evaluator), which supports a unified interface for evaluation across tens of benchmarks.
## Dataset Characterization:
** Data Collection Method<br>
* [Hybrid: Human, Synthetic, Automated]<br>
** Labeling Method<br>
* [Human] <br>
## Dataset Format:
Text.
## Dataset Quantification:
40 records
Each record contains the following fields:
- ID: Unique identifier for each sample
- Domain: Chemistry PhD / Physics PhD / Finance MBA / Consulting MBA
- Prompt: Instruction for the Large Language Model (LLM)
- Rubrics: 15-59 unique criterion used to assess the final model output
- Model Responses: 3 responses from OpenAI o3 / xAI Grok4 / DeepSeek R1-0528
Some portions of this dataset were created with Grok.
Total Storage: 1 MB.
## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
## Citation:
If you found ProfBench helpful, please consider citing the below:
```
@misc{wang2025profbenchmultidomainrubricsrequiring,
title={ProfBench: Multi-Domain Rubrics requiring Professional Knowledge to Answer and Judge},
author={Zhilin Wang and Jaehun Jung and Ximing Lu and Shizhe Diao and Ellie Evans and Jiaqi Zeng and Pavlo Molchanov and Yejin Choi and Jan Kautz and Yi Dong},
year={2025},
eprint={2510.18941},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.18941},
}
```
</details>
|