# super_glue by Amanpreet Singh: AI Training Dataset
Source: https://savrn.com/datasets/super-glue
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Dataset Card

SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after GLUE with a new set of more difficult language understanding tasks, improved resources, and a new public leaderboard. An example of 'test' looks as follows. An example of 'test' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. The data fields are the same among all splits. - sentence1: a string feature. - sentence2: a string feature. - idx: a int32 feature. - label: a classification label, with possible values including entailment (0), notentailment (1). - premise: a string feature. - hypothesis: a string feature. - idx: a…

Excerpt from the card by Amanpreet Singh, licensed other.

## Structure

### axb 1,104 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 1,104 | 243.8 KB |

sentence1stringsentence2stringidxint32labelClassLabel

### axg 356 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 356 | 53.4 KB |

premisestringhypothesisstringidxint32labelClassLabel

### boolq 15,942 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 9,427 | 6.2 MB |
| validation | 3,270 | 2.1 MB |
| test | 3,245 | 2.0 MB |

questionstringpassagestringidxint32labelClassLabel

### cb 556 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 250 | 87.1 KB |
| validation | 56 | 21.9 KB |
| test | 250 | 93.6 KB |

premisestringhypothesisstringidxint32labelClassLabel

### copa 1,000 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 400 | 49.3 KB |
| validation | 100 | 12.5 KB |
| test | 500 | 60.0 KB |

premisestringchoice1stringchoice2stringquestionstringidxint32labelClassLabel

### multirc 41,784 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 27,243 | 50.9 MB |
| validation | 4,848 | 7.8 MB |
| test | 9,693 | 15.7 MB |

paragraphstringquestionstringanswerstringidxvaluelabelClassLabel

### record 120,730 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 100,730 | 179.2 MB |
| validation | 10,000 | 17.6 MB |
| test | 10,000 | 17.2 MB |

passagestringquerystringentitiesSequenceentity_spansSequenceanswersSequenceidxvalue

### rte 5,767 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 2,490 | 857.0 KB |
| validation | 277 | 90.8 KB |
| test | 3,000 | 724.4 KB |

premisestringhypothesisstringidxint32labelClassLabel

### wic 7,466 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 5,428 | 661.9 KB |
| validation | 638 | 82.3 KB |
| test | 1,400 | 179.0 KB |

wordstringsentence1stringsentence2stringstart1int32start2int32end1int32end2int32idxint32labelClassLabel

### wsc 804 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 554 | 89.6 KB |
| validation | 104 | 21.6 KB |
| test | 146 | 31.5 KB |

textstringspan1_indexint32span2_indexint32span1_textstringspan2_textstringidxint32labelClassLabel

### wsc.fixed 804 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 554 | 89.6 KB |
| validation | 104 | 21.6 KB |
| test | 146 | 31.5 KB |

textstringspan1_indexint32span2_indexint32span1_textstringspan2_textstringidxint32labelClassLabel

## Details

Repository

aps/super_glue

Publisher

Amanpreet Singh

Task category

Text classification

Tags

superglue, NLU, natural language understanding

Size category

10K<n<100K

Languages

en

Revision

3de24cf8022e94f4ee4b9d55a6f539891524d646

Last updated

2025-05-16

## SAVRN's Notes on super_glue

SuperGLUE carries 196,313 rows across 11 configs in 102.5 MB, license listed as other, 685.9k monthly downloads, under the super_glue name with NLU tags. BoolQ is the biggest config at 15,942 rows; the diagnostic AX-b and AX-g sets are 1,104 and 356 rows. Our page shows two inbound models, both DeBERTa long-NLI variants: this is benchmark territory, not training territory.

The publisher line names an individual mirror, Amanpreet Singh's aps/super_glue, rather than the benchmark consortium, so pin the revision, last updated 2025-05-16, if you need reproducibility. Same play as GLUE: train on the larger configs' train splits, never on the eval sets, and read the per-task terms since other bundles several licenses. Expect modern base models to post near-ceiling scores.

SAVRN Research

## Files

31 files, 102.5 MB in total.

Data29 files · 102.5 MB

Documentation1 file · 19.7 KB

Repository1 file · 1.2 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| axb/test-00000-of-00001.parquet | Data | 80.9 KB | 655cadf2b23e |
| axg/test-00000-of-00001.parquet | Data | 14.3 KB | 4ed3eddc7286 |
| boolq/test-00000-of-00001.parquet | Data | 1.3 MB | 95a74cd7c8fd |
| boolq/train-00000-of-00001.parquet | Data | 3.8 MB | 41e2ce07ac0d |
| boolq/validation-00000-of-00001.parquet | Data | 1.3 MB | f46438b8cce0 |
| cb/test-00000-of-00001.parquet | Data | 63.5 KB | e9b1164fed6c |
| cb/train-00000-of-00001.parquet | Data | 58.0 KB | a338d1276fd0 |
| cb/validation-00000-of-00001.parquet | Data | 18.0 KB | 36cf71dba8a3 |
| copa/test-00000-of-00001.parquet | Data | 40.2 KB | 7d7d6e12adc0 |
| copa/train-00000-of-00001.parquet | Data | 33.9 KB | 8e243c3a2428 |
| copa/validation-00000-of-00001.parquet | Data | 12.0 KB | 247a7f586c5c |
| multirc/test-00000-of-00001.parquet | Data | 580.9 KB | 465107b93b83 |
| multirc/train-00000-of-00001.parquet | Data | 1.7 MB | ced6d731a335 |
| multirc/validation-00000-of-00001.parquet | Data | 306.4 KB | 5eb64893afc5 |
| record/test-00000-of-00001.parquet | Data | 8.0 MB | 6a91cd50dbdf |
| record/train-00000-of-00001.parquet | Data | 75.1 MB | 20ae20deb498 |
| record/validation-00000-of-00001.parquet | Data | 8.1 MB | 3662aa327fec |
| rte/test-00000-of-00001.parquet | Data | 622.0 KB | 5a4c57d7b81e |
| rte/train-00000-of-00001.parquet | Data | 586.2 KB | 16c83717da2e |
| rte/validation-00000-of-00001.parquet | Data | 69.8 KB | b3740aac8052 |
| wic/test-00000-of-00001.parquet | Data | 126.8 KB | e77d12183550 |
| wic/train-00000-of-00001.parquet | Data | 410.3 KB | d6d2231730f5 |
| wic/validation-00000-of-00001.parquet | Data | 60.1 KB | 22cc5df62bf7 |
| wsc.fixed/test-00000-of-00001.parquet | Data | 12.2 KB | 5308e3ca6543 |
| wsc.fixed/train-00000-of-00001.parquet | Data | 27.3 KB | 13086042694c |
| wsc.fixed/validation-00000-of-00001.parquet | Data | 10.2 KB | aa64cc8f27c7 |
| wsc/test-00000-of-00001.parquet | Data | 12.2 KB | 4a39c4f576a4 |
| wsc/train-00000-of-00001.parquet | Data | 27.2 KB | a1fb4be43c74 |
| wsc/validation-00000-of-00001.parquet | Data | 10.2 KB | 1459bd736725 |
| README.md | Documentation | 19.7 KB | — |
| .gitattributes | Repository | 1.2 KB | — |

## License and Download

License

other

Access

No access gate

[Download from Amanpreet Singh](https://huggingface.co/datasets/aps/super_glue)

Released by Amanpreet Singh through its official repository on Hugging Face.

## Models Trained on This Dataset

- Trained on (disclosed)[deberta-small-long-nli](https://savrn.com/models/deberta-small-long-nli)
- Trained on (disclosed)[deberta-base-long-nli](https://savrn.com/models/deberta-base-long-nli)

## Amanpreet Singh

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

## 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.
- [Hugging Face record](https://huggingface.co/datasets/aps/super_glue)
- [How the hub is built](https://savrn.com/model-hub/methodology)