# winogrande by Ai2: AI Training Dataset Source: https://savrn.com/datasets/winogrande Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Dataset Card WinoGrande is a new collection of 44k problems, inspired by Winograd Schema Challenge (Levesque, Davis, and Morgenstern 2011), but adjusted to improve the scale and robustness against the dataset-specific bias. Formulated as a fill-in-a-blank task with binary options, the goal is to choose the right option for a given sentence which requires commonsense reasoning. An example of 'train' looks as follows. An example of 'validation' looks as follows. An example of 'validation' looks as follows. An example of 'validation' looks as follows. An example of 'train' looks as follows. The data fields are the same among all splits. - sentence: a string feature. - option1: a string feature. - option2… Excerpt from the card by Ai2. ## Structure ### winogrande_debiased 12,282 rows | Split | Rows | Size | | --- | --- | --- | | train | 9,248 | 1.1 MB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ### winogrande_l 13,268 rows | Split | Rows | Size | | --- | --- | --- | | train | 10,234 | 1.2 MB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ### winogrande_m 5,592 rows | Split | Rows | Size | | --- | --- | --- | | train | 2,558 | 289.4 KB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ### winogrande_s 3,674 rows | Split | Rows | Size | | --- | --- | --- | | train | 640 | 72.6 KB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ### winogrande_xl 43,432 rows | Split | Rows | Size | | --- | --- | --- | | train | 40,398 | 4.5 MB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ### winogrande_xs 3,194 rows | Split | Rows | Size | | --- | --- | --- | | train | 160 | 19.0 KB | | test | 1,767 | 204.5 KB | | validation | 1,267 | 147.4 KB | sentencestringoption1stringoption2stringanswerstring ## Details Repository allenai/winogrande Publisher Ai2 Task category Not stated by the source Tags Not stated by the source Size category Not stated by the source Languages en Revision 01e74176c63542e6b0bcb004dcdea22d94fb67b5 Last updated 2025-07-11 ## SAVRN's Notes on winogrande WinoGrande is Ai2's scaled-up Winograd challenge: 81,442 rows across six configurations — xs 3,194 up to xl 43,432, plus the 12,282-row debiased set — in 4.6 MB and 20 files, downloaded 404.4k times a month. Each problem is a fill-in-the-blank sentence with two candidate answers that demands commonsense resolution. The card calls it a collection of 44k problems adjusted for scale and robustness against dataset-specific bias, so treat the sizes as sampling tiers of one design. It is benchmark material: train on the official train splits, never on validation, and expect any published number to be compared against a paper trail going back years. The debiased configuration is the split to trust if you fine-tune, precisely because the card flags bias-robustness as the design goal. One real gap: the license field reads 'Not stated by the source' on this record, so a commercial training run needs terms confirmed with Ai2 before weights ship. SAVRN Research ## Files 20 files, 4.6 MB in total. Data18 files · 4.6 MB Documentation1 file · 11.2 KB Repository1 file · 1.2 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | winogrande_debiased/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_debiased/train-00000-of-00001.parquet | Data | 616.8 KB | b69a6e9c13a6 | | winogrande_debiased/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | winogrande_l/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_l/train-00000-of-00001.parquet | Data | 529.5 KB | d6f5c0c5d288 | | winogrande_l/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | winogrande_m/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_m/train-00000-of-00001.parquet | Data | 133.8 KB | a3be4246db58 | | winogrande_m/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | winogrande_s/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_s/train-00000-of-00001.parquet | Data | 34.8 KB | ce089cd597b8 | | winogrande_s/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | winogrande_xl/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_xl/train-00000-of-00001.parquet | Data | 2.1 MB | 5d8c38ad12b9 | | winogrande_xl/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | winogrande_xs/test-00000-of-00001.parquet | Data | 117.7 KB | 100885ec03a6 | | winogrande_xs/train-00000-of-00001.parquet | Data | 11.7 KB | a4698ccae8c0 | | winogrande_xs/validation-00000-of-00001.parquet | Data | 85.9 KB | f9d914d1818c | | README.md | Documentation | 11.2 KB | — | | .gitattributes | Repository | 1.2 KB | — | ## License and Download License Not stated by the source Access No access gate [Download from Ai2](https://huggingface.co/datasets/allenai/winogrande) Released by Ai2 through its official repository on Hugging Face. ## Models Trained on This Dataset - Trained on (disclosed)[shingi-27b](https://savrn.com/models/shingi-27b) ## 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/winogrande) - [How the hub is built](https://savrn.com/model-hub/methodology)