# sciq by Ai2: AI Training Dataset
Source: https://savrn.com/datasets/sciq
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Dataset Card

The SciQ dataset contains 13,679 crowdsourced science exam questions about Physics, Chemistry and Biology, among others. The questions are in multiple-choice format with 4 answer options each. For the majority of the questions, an additional paragraph with supporting evidence for the correct answer is provided. An example of 'train' looks as follows. The data fields are the same among all splits. - question: a string feature. - distractor3: a string feature. - distractor1: a string feature. - distractor2: a string feature. - correctanswer: a string feature. - support: a string feature. The dataset is licensed under the Creative Commons Attribution-NonCommercial 3.0 Unported License. Thanks…

Excerpt from the card by Ai2, licensed cc-by-nc-3.0.

## Structure

### default 13,679 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 11,679 | 6.6 MB |
| validation | 1,000 | 554.1 KB |
| test | 1,000 | 563.9 KB |

questionstringdistractor3stringdistractor1stringdistractor2stringcorrect_answerstringsupportstring

## Details

Repository

allenai/sciq

Publisher

Ai2

Task category

Question answering

Tags

Not stated by the source

Size category

10K<n<100K

Languages

en

Revision

2c94ad3e1aafab77146f384e23536f97a4849815

Last updated

2024-01-04

## SAVRN's Notes on sciq

SciQ is 13,679 crowdsourced science exam questions — physics, chemistry, biology — in 4.7 MB, split 11,679 train, 1,000 validation and 1,000 test, downloaded 392.5k times a month. Every question is multiple-choice with four options, and for most items a support paragraph evidencing the correct answer is included, which makes the set doubly useful: closed-book QA eval, or a ready retrieval/RAG testbed. At 4.7 MB and five files it costs nothing to keep around.

The license is CC BY-NC 3.0: fine for internal research, off-limits for commercial model training without a separate deal — the NC clause is the whole story here. Also treat the test split as burned; this set is old, small and heavily reported, so a score on it says little about general science ability. Train on the train split only if your use is non-commercial, and otherwise pick a permissively licensed alternative before committing.

SAVRN Research

## Files

5 files, 4.7 MB in total.

Data3 files · 4.7 MB

Documentation1 file · 7.0 KB

Repository1 file · 1.2 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| data/test-00000-of-00001.parquet | Data | 342.8 KB | 3a719356a29b |
| data/train-00000-of-00001.parquet | Data | 4.0 MB | 196443609540 |
| data/validation-00000-of-00001.parquet | Data | 338.5 KB | 455dd9f1d725 |
| README.md | Documentation | 7.0 KB | — |
| .gitattributes | Repository | 1.2 KB | — |

## License and Download

License

cc-by-nc-3.0

Access

No access gate

[Download from Ai2](https://huggingface.co/datasets/allenai/sciq)

Released by Ai2 through its official repository on Hugging Face.

## Ai2

[All models and datasets](https://savrn.com/model-publishers/allenai)

## Explore More

- [The dataset directory](https://savrn.com/datasets)
- [The model directory](https://savrn.com/models)
- [Research papers](https://savrn.com/papers)

## Source

- Repository metadata, read 2026-10-19.
- [Ai2 website](https://allenai.org/)
- [Hugging Face record](https://huggingface.co/datasets/allenai/sciq)
- [How the hub is built](https://savrn.com/model-hub/methodology)