README.md

---
pretty_name: JevalKit Datasets
license: other
license_name: upstream-dataset-specific-licenses
license_link: https://huggingface.co/datasets/Valen-Team/JevalKit_Datasets/blob/main/README.md#sources-and-licenses
language: [en, zh, de, tr]
tags: [jev, decision-making, benchmark, evaluation, jevalkit]
configs:
- config_name: VisualDecisionBench
  data_files: [{split: test, path: datasets/VisualDecisionBench.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: VisualDecisionBench_Image
  data_files: [{split: test, path: datasets/VisualDecisionBench_Image.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: VisualDecisionBench_Video
  data_files: [{split: test, path: datasets/VisualDecisionBench_Video.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: JevBench
  data_files: [{split: test, path: datasets/JevBench.tsv}]
  sep: "\t"
  keep_default_na: false
  default: true
- config_name: DecisionIndex
  data_files: [{split: test, path: datasets/DecisionIndex.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: TypedDecisions
  data_files: [{split: test, path: datasets/TypedDecisions.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: DecisionBench
  data_files: [{split: test, path: datasets/DecisionBench.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: ProceduralTypedDecisions
  data_files: [{split: test, path: datasets/ProceduralTypedDecisions.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: ZhDecisionBench
  data_files: [{split: test, path: datasets/ZhDecisionBench.tsv}]
  sep: "\t"
  keep_default_na: false
- config_name: JevAltBench
  data_files: [{split: test, path: datasets/JevAltBench.tsv}]
  sep: "\t"
  keep_default_na: false
---

# JevalKit Datasets

TSV evaluation datasets for [JevalKit](https://github.com/Liuziyu77/JevalKit): **10 views across 8 benchmark families**.

| Dataset / upstream source | Records | Questions | License |
|---|---:|---:|---|
| [VisualDecisionBench](https://huggingface.co/datasets/Valen-Team/VisualDecisionBench) | 1,696 | 9,143 | Per source dataset |
| [VisualDecisionBench_Image](https://huggingface.co/datasets/Valen-Team/VisualDecisionBench) | 1,402 | 2,000 | Per source dataset |
| [VisualDecisionBench_Video](https://huggingface.co/datasets/Valen-Team/VisualDecisionBench) | 294 | 7,143 | Per source dataset |
| [JevBench](https://github.com/fstandhartinger/jevbench) | 231 | 231 | MIT |
| [DecisionIndex](https://github.com/apolinario/decision-index) | 140,178 | 312,320 | Per source dataset |
| [TypedDecisions](https://huggingface.co/datasets/LocalLLaMA/typed-decisions) | 400 | 2,000 | Apache-2.0 |
| [DecisionBench](https://huggingface.co/datasets/Hanno-Labs/decision-bench) | 23,900 | 23,900 | Source-specific |
| [ProceduralTypedDecisions](https://huggingface.co/datasets/tasksource/procedural-typed-decisions) | 18,000 | 57,016 | Apache-2.0 |
| [ZhDecisionBench](https://huggingface.co/datasets/CodyQin/zh-decision-bench) | 378 | 467 | CC-BY-4.0 |
| [JevAltBench](https://huggingface.co/datasets/mertkayacs/jevalt-bench) | 3,116 | 8,434 | Per-record licenses |

`VisualDecisionBench` combines the Image and Video views; these files overlap. All configurations expose a `test` evaluation split.

## Usage

```python
import json
from datasets import load_dataset

data = load_dataset("Valen-Team/JevalKit_Datasets", "JevBench", split="test")
request = json.loads(data[0]["request"])
targets = json.loads(data[0]["targets"])
```

Columns: `index`, `request`, `targets`, `meta`, `assets`. The last four contain JSON strings; each row holds one shared state and its questions.

Visual media are available from [VisualDecisionBench](https://huggingface.co/datasets/Valen-Team/VisualDecisionBench); asset paths remain in the TSVs.

Download the ready-to-use TSVs with `hf download Valen-Team/JevalKit_Datasets --repo-type dataset --include "datasets/*.tsv" --local-dir "$JEVALKIT_DATA_ROOT"`. JevalKit reads pinned sources, protocols and checksums from the manifests bundled in its [benchmark directories](https://github.com/Valen-Team/JevalKit/tree/main/jevalkit/datasets); no local manifest or conversion is required. Visual media and Decision Index scoring inputs have separate downloads in the [usage guide](https://github.com/Valen-Team/JevalKit/blob/main/docs/quickstart.md#download-data).

## Sources and licenses

Dataset names above link to their upstream sources. Each dataset retains its original terms and attribution; record-level provenance remains in `meta`. ZhDecisionBench credits Amazon MASSIVE and CodyQin under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/).

<details>
<summary>JevBench MIT notice</summary>

MIT License

Copyright (c) 2026 Florian Standhartinger and contributors

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

</details>