BoolQ is a question answering dataset for yes/no questions containing 15942 examples. These questions are naturally occurring ---they are generated in unprompted and unconstrained settings.
Dataset Card
By Google, published under cc-by-sa-3.0, revision 35b264d03638.
Dataset Card for Boolq
Table of Contents
- Dataset Description
- Dataset Summary
- Supported Tasks and Leaderboards
- Languages
- Dataset Structure
- Data Instances
- Data Fields
- Data Splits
- Dataset Creation
- Curation Rationale
- Source Data
- Annotations
- Personal and Sensitive Information
- Considerations for Using the Data
- Social Impact of Dataset
- Discussion of Biases
- Other Known Limitations
- Additional Information
- Dataset Curators
- Licensing Information
- Citation Information
- Contributions
Dataset Description
- Homepage: More Information Needed
- Repository: https://github.com/google-research-datasets/boolean-questions
- Paper: https://arxiv.org/abs/1905.10044
- Point of Contact: More Information Needed
- Size of downloaded dataset files: 8.77 MB
- Size of the generated dataset: 7.83 MB
- Total amount of disk used: 16.59 MB
Dataset Summary
BoolQ is a question answering dataset for yes/no questions containing 15942 examples. These questions are naturally occurring ---they are generated in unprompted and unconstrained settings. Each example is a triplet of (question, passage, answer), with the title of the page as optional additional context. The text-pair classification setup is similar to existing natural language inference tasks.
Supported Tasks and Leaderboards
Structure
default 12,697 rows
| Split | Rows | Size |
|---|---|---|
| train | 9,427 | 5.8 MB |
| validation | 3,270 | 2.0 MB |
Details
- Repository
- google/boolq
- Publisher
- Task category
- Text classification
- Tags
- Not stated by the source
- Size category
- 10K<n<100K
- Languages
- en
- Revision
- 35b264d03638db9f4ce671b711558bf7ff0f80d5
- Last updated
- 2024-01-22
SAVRN's Notes on boolq
BoolQ ships 12,697 yes/no reading-comprehension rows — 9,427 train, 3,270 validation — in a 5.0 MB download, though the card text claims 15,942 examples; the structure table is the number to plan against. The questions are naturally occurring search-style queries paired with Wikipedia passages, which makes it a cheap classification benchmark but also means overlap with any Wikipedia-heavy pretraining corpus. The inbound models on our page are all small decision or fine-tuned classifiers — Jev-Style-Qwen 2B variants, kev-4b, kev-8b, dev-0.4b — which is the right scale of use.
License is CC BY-SA 3.0, an unusual pick for a Google dataset: attribution is easy, but the share-alike clause can attach to adapted material in a commercial product, so have counsel look before fine-tuning customer-facing models on it. At 5.0 MB and 206.7k monthly downloads it is firmly a benchmark, not training fuel. Use it to sanity-check boolean reasoning; do not expect a 12k-row set of 2019-era queries to move model capability.
Files
4 files, 5.0 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| data/train-00000-of-00001.parquet | Data | 3.7 MB | 4f028e992c0b |
| data/validation-00000-of-00001.parquet | Data | 1.3 MB | 52355d11524b |
| README.md | Documentation | 6.6 KB | — |
| .gitattributes | Repository | 1.2 KB | — |
License and Download
- License
- cc-by-sa-3.0
- Access
- No access gate
Released by Google through its official repository on Hugging Face.
Models Trained on This Dataset
- Trained on (disclosed)Jev-Style-Qwen3.5-2B-Decision-GGUF
- Trained on (disclosed)Jev-Style-Qwen3.5-2B-Decision-MLX-bf16
- Trained on (disclosed)hopper
- Trained on (disclosed)kev-4b
- Trained on (disclosed)kev-8b
- Trained on (disclosed)dev-0.4b
- Trained on (disclosed)bobcat-flash-1.2