Datasets:

permitt
/

superglue-hr

license: cc-by-4.0
language:
  - hr
multilinguality: monolingual
pretty_name: BalkanBench SuperGLUE - Croatian
size_categories:
  - 10K<n<100K
source_datasets:
  - extended|super_glue
task_categories:
  - text-classification
  - multiple-choice
  - question-answering
tags:
  - balkanbench
  - bcms
  - croatian
  - superglue
  - nlu
  - benchmark
configs:
  - config_name: boolq
    data_files:
      - split: train
        path: boolq/train-00000-of-00001.parquet
      - split: validation
        path: boolq/validation-00000-of-00001.parquet
  - config_name: cb
    data_files:
      - split: train
        path: cb/train-00000-of-00001.parquet
      - split: validation
        path: cb/validation-00000-of-00001.parquet
  - config_name: copa
    data_files:
      - split: train
        path: copa/train-00000-of-00001.parquet
      - split: validation
        path: copa/validation-00000-of-00001.parquet
  - config_name: multirc
    data_files:
      - split: train
        path: multirc/train-00000-of-00001.parquet
      - split: validation
        path: multirc/validation-00000-of-00001.parquet
  - config_name: rte
    data_files:
      - split: train
        path: rte/train-00000-of-00001.parquet
      - split: validation
        path: rte/validation-00000-of-00001.parquet

BalkanBench SuperGLUE - Croatian

Part of BalkanBench - the open, reproducible benchmark for language models across Serbian, Croatian, Montenegrin, and Bosnian (BCMS). Live leaderboard at https://balkanbench.com/leaderboard. Background: Release of BalkanBench - the vision behind it (Medium, 2026-04-27).

This is the Croatian SuperGLUE released preview track of BalkanBench v0.1. Croatian publishes alongside Serbian (the official frozen track) and Montenegrin as a 5-task preview: same scoring contract, same private test labels, no WSC adaptation yet.

What's inside

5 task configs, full train + validation + public test splits (train and validation labelled here; the test split lives only in the gated sibling repo, see below):

Config	What it tests	Train	Validation	Test
boolq	Yes/no question answering over passages	9,427	3,270	3,245
cb	3-way textual entailment (entail / contradict / neutral)	250	56	250
copa	Causal reasoning between two alternatives	400	100	500
multirc	Multi-sentence reading comprehension (multiple correct)	27,243	4,848	9,693
rte	Binary textual entailment	2,490	277	3,000
TOTAL		39,810	8,551	16,688

Total: 65,049 items. WSC and the AX-b/AX-g diagnostics are Serbian-only in v0.1 - they may be added in a later release once a Croatian adaptation is available.

Test split is gated

The test split for every config lives only in the gated sibling repo (both inputs and labels): permitt/superglue-hr-private, and are accessed only by the official scoring pipeline. To get a leaderboard score you generate predictions on the public test inputs with balkanbench predict, then submit through the submission flow.

Loading

from datasets import load_dataset

boolq = load_dataset("permitt/superglue-hr", "boolq")
print(boolq["train"][0])

# All ranked tasks
for cfg in ["boolq", "cb", "copa", "rte", "multirc"]:
    ds = load_dataset("permitt/superglue-hr", cfg)
    print(cfg, {sp: len(ds[sp]) for sp in ds})

Pin a revision for reproducible runs:

load_dataset("permitt/superglue-hr", "boolq", revision="v0.1.0-data")

Schema

Each example carries an integer row id idx and the task-specific input fields plus a label column. For cb:

{
    "idx": 0,
    "premise": "Polly je morala brzo razmišljati. ...",
    "hypothesis": "Polly nije bila iskusna oceanska jedriličarka",
    "label": 0,  # 0=entailment, 1=contradiction, 2=neutral
}

Per-task field lists are documented in the BalkanBench source repo.

Methodology

Translation: original SuperGLUE English -> Croatian via translation + human verification by native speakers.
Frozen splits: train/validation/test row counts and IDs pinned at tag v0.1.0-data.
Test labels stay private: see the gated sibling repo above.

Citation

If you use this dataset, please cite the Serbian SuperGLUE paper (which introduces the BCMS adaptation methodology used here), the original SuperGLUE paper, and the BalkanBench release:

@inproceedings{perovic-mihajlov-2026-serbian,
    title = "{S}erbian {S}uper{GLUE}: Towards an Evaluation Benchmark for {S}outh {S}lavic Language Models",
    author = "Perovic, Mitar and Mihajlov, Teodora",
    editor = "Hettiarachchi, Hansi and Ranasinghe, Tharindu and Plum, Alistair and
              Rayson, Paul and Mitkov, Ruslan and Gaber, Mohamed and Premasiri, Damith and
              Tan, Fiona Anting and Uyangodage, Lasitha",
    booktitle = "Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.loreslm-1.30/",
    doi = "10.18653/v1/2026.loreslm-1.30",
    pages = "347--361",
    ISBN = "979-8-89176-377-7"
}

@inproceedings{wang2019superglue,
  title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},
  author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and
          Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R.},
  booktitle={NeurIPS},
  year={2019}
}

@misc{balkanbench2026superglue_hr,
  title={BalkanBench SuperGLUE-HR: Croatian SuperGLUE for the BCMS evaluation suite},
  author={Perović, Mitar and contributors},
  year={2026},
  howpublished={\url{https://huggingface.co/datasets/permitt/superglue-hr}},
  note={Part of BalkanBench v0.1, \url{https://balkanbench.com}}
}

License

CC-BY-4.0 (matches the original SuperGLUE source license).

Sponsor

Compute for the official v0.1 evaluation is sponsored by Recrewty.