---
pretty_name: EuroBench parallel (reshuffled)
license: other
license_name: mixed-see-sources
task_categories:
- multiple-choice
- question-answering
configs:
- config_name: mmlu_cs
  data_files:
  - split: test
    path: mmlu/cs/test.parquet
  - split: dev
    path: mmlu/cs/dev.parquet
- config_name: mmlu_da
  data_files:
  - split: test
    path: mmlu/da/test.parquet
  - split: dev
    path: mmlu/da/dev.parquet
- config_name: mmlu_de
  data_files:
  - split: test
    path: mmlu/de/test.parquet
  - split: dev
    path: mmlu/de/dev.parquet
- config_name: mmlu_el
  data_files:
  - split: test
    path: mmlu/el/test.parquet
  - split: dev
    path: mmlu/el/dev.parquet
- config_name: mmlu_en
  data_files:
  - split: test
    path: mmlu/en/test.parquet
  - split: dev
    path: mmlu/en/dev.parquet
- config_name: mmlu_es
  data_files:
  - split: test
    path: mmlu/es/test.parquet
  - split: dev
    path: mmlu/es/dev.parquet
- config_name: mmlu_et
  data_files:
  - split: test
    path: mmlu/et/test.parquet
  - split: dev
    path: mmlu/et/dev.parquet
- config_name: mmlu_fi
  data_files:
  - split: test
    path: mmlu/fi/test.parquet
  - split: dev
    path: mmlu/fi/dev.parquet
- config_name: mmlu_fr
  data_files:
  - split: test
    path: mmlu/fr/test.parquet
  - split: dev
    path: mmlu/fr/dev.parquet
- config_name: mmlu_ga
  data_files:
  - split: test
    path: mmlu/ga/test.parquet
  - split: dev
    path: mmlu/ga/dev.parquet
- config_name: mmlu_hr
  data_files:
  - split: test
    path: mmlu/hr/test.parquet
  - split: dev
    path: mmlu/hr/dev.parquet
- config_name: mmlu_hu
  data_files:
  - split: test
    path: mmlu/hu/test.parquet
  - split: dev
    path: mmlu/hu/dev.parquet
- config_name: mmlu_it
  data_files:
  - split: test
    path: mmlu/it/test.parquet
  - split: dev
    path: mmlu/it/dev.parquet
- config_name: mmlu_lt
  data_files:
  - split: test
    path: mmlu/lt/test.parquet
  - split: dev
    path: mmlu/lt/dev.parquet
- config_name: mmlu_lv
  data_files:
  - split: test
    path: mmlu/lv/test.parquet
  - split: dev
    path: mmlu/lv/dev.parquet
- config_name: mmlu_nl
  data_files:
  - split: test
    path: mmlu/nl/test.parquet
  - split: dev
    path: mmlu/nl/dev.parquet
- config_name: mmlu_pl
  data_files:
  - split: test
    path: mmlu/pl/test.parquet
  - split: dev
    path: mmlu/pl/dev.parquet
- config_name: mmlu_pt
  data_files:
  - split: test
    path: mmlu/pt/test.parquet
  - split: dev
    path: mmlu/pt/dev.parquet
- config_name: mmlu_ro
  data_files:
  - split: test
    path: mmlu/ro/test.parquet
  - split: dev
    path: mmlu/ro/dev.parquet
- config_name: mmlu_ru
  data_files:
  - split: test
    path: mmlu/ru/test.parquet
  - split: dev
    path: mmlu/ru/dev.parquet
- config_name: mmlu_sk
  data_files:
  - split: test
    path: mmlu/sk/test.parquet
  - split: dev
    path: mmlu/sk/dev.parquet
- config_name: mmlu_sl
  data_files:
  - split: test
    path: mmlu/sl/test.parquet
  - split: dev
    path: mmlu/sl/dev.parquet
- config_name: mmlu_sr
  data_files:
  - split: test
    path: mmlu/sr/test.parquet
  - split: dev
    path: mmlu/sr/dev.parquet
- config_name: mmlu_sv
  data_files:
  - split: test
    path: mmlu/sv/test.parquet
  - split: dev
    path: mmlu/sv/dev.parquet
- config_name: mmlu_tr
  data_files:
  - split: test
    path: mmlu/tr/test.parquet
  - split: dev
    path: mmlu/tr/dev.parquet
- config_name: mmlu_uk
  data_files:
  - split: test
    path: mmlu/uk/test.parquet
  - split: dev
    path: mmlu/uk/dev.parquet
- config_name: belebele_en
  data_files:
  - split: test
    path: belebele/en/test.parquet
- config_name: belebele_hr
  data_files:
  - split: test
    path: belebele/hr/test.parquet
- config_name: belebele_el
  data_files:
  - split: test
    path: belebele/el/test.parquet
- config_name: belebele_it
  data_files:
  - split: test
    path: belebele/it/test.parquet
- config_name: belebele_mt
  data_files:
  - split: test
    path: belebele/mt/test.parquet
- config_name: belebele_pt
  data_files:
  - split: test
    path: belebele/pt/test.parquet
- config_name: belebele_sl
  data_files:
  - split: test
    path: belebele/sl/test.parquet
- config_name: belebele_es
  data_files:
  - split: test
    path: belebele/es/test.parquet
- config_name: belebele_cs
  data_files:
  - split: test
    path: belebele/cs/test.parquet
- config_name: belebele_de
  data_files:
  - split: test
    path: belebele/de/test.parquet
- config_name: belebele_hu
  data_files:
  - split: test
    path: belebele/hu/test.parquet
- config_name: belebele_pl
  data_files:
  - split: test
    path: belebele/pl/test.parquet
- config_name: belebele_sk
  data_files:
  - split: test
    path: belebele/sk/test.parquet
- config_name: belebele_nl
  data_files:
  - split: test
    path: belebele/nl/test.parquet
- config_name: belebele_fr
  data_files:
  - split: test
    path: belebele/fr/test.parquet
- config_name: belebele_da
  data_files:
  - split: test
    path: belebele/da/test.parquet
- config_name: belebele_fi
  data_files:
  - split: test
    path: belebele/fi/test.parquet
- config_name: belebele_sv
  data_files:
  - split: test
    path: belebele/sv/test.parquet
- config_name: belebele_bg
  data_files:
  - split: test
    path: belebele/bg/test.parquet
- config_name: belebele_et
  data_files:
  - split: test
    path: belebele/et/test.parquet
- config_name: belebele_lv
  data_files:
  - split: test
    path: belebele/lv/test.parquet
- config_name: belebele_lt
  data_files:
  - split: test
    path: belebele/lt/test.parquet
- config_name: belebele_ro
  data_files:
  - split: test
    path: belebele/ro/test.parquet
- config_name: belebele_ru
  data_files:
  - split: test
    path: belebele/ru/test.parquet
- config_name: belebele_sr
  data_files:
  - split: test
    path: belebele/sr/test.parquet
- config_name: belebele_tr
  data_files:
  - split: test
    path: belebele/tr/test.parquet
- config_name: belebele_uk
  data_files:
  - split: test
    path: belebele/uk/test.parquet
- config_name: arc_challenge_en
  data_files:
  - split: test
    path: arc_challenge/en/test.parquet
  - split: validation
    path: arc_challenge/en/validation.parquet
- config_name: arc_challenge_bg
  data_files:
  - split: test
    path: arc_challenge/bg/test.parquet
  - split: validation
    path: arc_challenge/bg/validation.parquet
- config_name: arc_challenge_cs
  data_files:
  - split: test
    path: arc_challenge/cs/test.parquet
  - split: validation
    path: arc_challenge/cs/validation.parquet
- config_name: arc_challenge_da
  data_files:
  - split: test
    path: arc_challenge/da/test.parquet
  - split: validation
    path: arc_challenge/da/validation.parquet
- config_name: arc_challenge_de
  data_files:
  - split: test
    path: arc_challenge/de/test.parquet
  - split: validation
    path: arc_challenge/de/validation.parquet
- config_name: arc_challenge_el
  data_files:
  - split: test
    path: arc_challenge/el/test.parquet
  - split: validation
    path: arc_challenge/el/validation.parquet
- config_name: arc_challenge_es
  data_files:
  - split: test
    path: arc_challenge/es/test.parquet
  - split: validation
    path: arc_challenge/es/validation.parquet
- config_name: arc_challenge_et
  data_files:
  - split: test
    path: arc_challenge/et/test.parquet
  - split: validation
    path: arc_challenge/et/validation.parquet
- config_name: arc_challenge_fi
  data_files:
  - split: test
    path: arc_challenge/fi/test.parquet
  - split: validation
    path: arc_challenge/fi/validation.parquet
- config_name: arc_challenge_fr
  data_files:
  - split: test
    path: arc_challenge/fr/test.parquet
  - split: validation
    path: arc_challenge/fr/validation.parquet
- config_name: arc_challenge_hu
  data_files:
  - split: test
    path: arc_challenge/hu/test.parquet
  - split: validation
    path: arc_challenge/hu/validation.parquet
- config_name: arc_challenge_it
  data_files:
  - split: test
    path: arc_challenge/it/test.parquet
  - split: validation
    path: arc_challenge/it/validation.parquet
- config_name: arc_challenge_lt
  data_files:
  - split: test
    path: arc_challenge/lt/test.parquet
  - split: validation
    path: arc_challenge/lt/validation.parquet
- config_name: arc_challenge_lv
  data_files:
  - split: test
    path: arc_challenge/lv/test.parquet
  - split: validation
    path: arc_challenge/lv/validation.parquet
- config_name: arc_challenge_nl
  data_files:
  - split: test
    path: arc_challenge/nl/test.parquet
  - split: validation
    path: arc_challenge/nl/validation.parquet
- config_name: arc_challenge_pl
  data_files:
  - split: test
    path: arc_challenge/pl/test.parquet
  - split: validation
    path: arc_challenge/pl/validation.parquet
- config_name: arc_challenge_pt
  data_files:
  - split: test
    path: arc_challenge/pt/test.parquet
  - split: validation
    path: arc_challenge/pt/validation.parquet
- config_name: arc_challenge_ro
  data_files:
  - split: test
    path: arc_challenge/ro/test.parquet
  - split: validation
    path: arc_challenge/ro/validation.parquet
- config_name: arc_challenge_sk
  data_files:
  - split: test
    path: arc_challenge/sk/test.parquet
  - split: validation
    path: arc_challenge/sk/validation.parquet
- config_name: arc_challenge_sl
  data_files:
  - split: test
    path: arc_challenge/sl/test.parquet
  - split: validation
    path: arc_challenge/sl/validation.parquet
- config_name: arc_challenge_sv
  data_files:
  - split: test
    path: arc_challenge/sv/test.parquet
  - split: validation
    path: arc_challenge/sv/validation.parquet
- config_name: arc_easy_en
  data_files:
  - split: test
    path: arc_easy/en/test.parquet
  - split: validation
    path: arc_easy/en/validation.parquet
- config_name: arc_easy_bg
  data_files:
  - split: test
    path: arc_easy/bg/test.parquet
  - split: validation
    path: arc_easy/bg/validation.parquet
- config_name: arc_easy_cs
  data_files:
  - split: test
    path: arc_easy/cs/test.parquet
  - split: validation
    path: arc_easy/cs/validation.parquet
- config_name: arc_easy_da
  data_files:
  - split: test
    path: arc_easy/da/test.parquet
  - split: validation
    path: arc_easy/da/validation.parquet
- config_name: arc_easy_de
  data_files:
  - split: test
    path: arc_easy/de/test.parquet
  - split: validation
    path: arc_easy/de/validation.parquet
- config_name: arc_easy_el
  data_files:
  - split: test
    path: arc_easy/el/test.parquet
  - split: validation
    path: arc_easy/el/validation.parquet
- config_name: arc_easy_es
  data_files:
  - split: test
    path: arc_easy/es/test.parquet
  - split: validation
    path: arc_easy/es/validation.parquet
- config_name: arc_easy_et
  data_files:
  - split: test
    path: arc_easy/et/test.parquet
  - split: validation
    path: arc_easy/et/validation.parquet
- config_name: arc_easy_fi
  data_files:
  - split: test
    path: arc_easy/fi/test.parquet
  - split: validation
    path: arc_easy/fi/validation.parquet
- config_name: arc_easy_fr
  data_files:
  - split: test
    path: arc_easy/fr/test.parquet
  - split: validation
    path: arc_easy/fr/validation.parquet
- config_name: arc_easy_hu
  data_files:
  - split: test
    path: arc_easy/hu/test.parquet
  - split: validation
    path: arc_easy/hu/validation.parquet
- config_name: arc_easy_it
  data_files:
  - split: test
    path: arc_easy/it/test.parquet
  - split: validation
    path: arc_easy/it/validation.parquet
- config_name: arc_easy_lt
  data_files:
  - split: test
    path: arc_easy/lt/test.parquet
  - split: validation
    path: arc_easy/lt/validation.parquet
- config_name: arc_easy_lv
  data_files:
  - split: test
    path: arc_easy/lv/test.parquet
  - split: validation
    path: arc_easy/lv/validation.parquet
- config_name: arc_easy_nl
  data_files:
  - split: test
    path: arc_easy/nl/test.parquet
  - split: validation
    path: arc_easy/nl/validation.parquet
- config_name: arc_easy_pl
  data_files:
  - split: test
    path: arc_easy/pl/test.parquet
  - split: validation
    path: arc_easy/pl/validation.parquet
- config_name: arc_easy_pt
  data_files:
  - split: test
    path: arc_easy/pt/test.parquet
  - split: validation
    path: arc_easy/pt/validation.parquet
- config_name: arc_easy_ro
  data_files:
  - split: test
    path: arc_easy/ro/test.parquet
  - split: validation
    path: arc_easy/ro/validation.parquet
- config_name: arc_easy_sk
  data_files:
  - split: test
    path: arc_easy/sk/test.parquet
  - split: validation
    path: arc_easy/sk/validation.parquet
- config_name: arc_easy_sl
  data_files:
  - split: test
    path: arc_easy/sl/test.parquet
  - split: validation
    path: arc_easy/sl/validation.parquet
- config_name: arc_easy_sv
  data_files:
  - split: test
    path: arc_easy/sv/test.parquet
  - split: validation
    path: arc_easy/sv/validation.parquet
- config_name: polymath_en
  data_files:
  - split: test
    path: polymath/en/test.parquet
- config_name: polymath_de
  data_files:
  - split: test
    path: polymath/de/test.parquet
- config_name: polymath_es
  data_files:
  - split: test
    path: polymath/es/test.parquet
- config_name: polymath_fr
  data_files:
  - split: test
    path: polymath/fr/test.parquet
- config_name: polymath_it
  data_files:
  - split: test
    path: polymath/it/test.parquet
- config_name: polymath_pt
  data_files:
  - split: test
    path: polymath/pt/test.parquet
- config_name: polymath_ru
  data_files:
  - split: test
    path: polymath/ru/test.parquet
---
# EuroBench Parallel

This repo contains several multilingual parallel benchmarks covering European languages ready to be used with `lm-eval-harness`. 

## Collection
As of **October, 2026**, it includes
- MMLU-mix, our own build, cleaned and refined from:
    - `cais/mmlu`
    - `openai/mmmlu`
    - `CohereLabs/Global-MMLU`
    - `edinburgh-dawg/mmlu-redux-2`
    - `collections/INSAIT-Institute/multilingual-benchmarks`
    - `Eurolingua/mmlux`
    - `EC-DGT-AI/EU-MMLU` (Croatian and Irish only)

  Croatian (`mmlu_hr`, 642 items) and Irish (`mmlu_ga`, 287 items) are partial: they have only the items their source covers, so their scores are not comparable with the other languages' (5,315 items) and should be reported separately.
- BELEBELE, copied from `facebook/belebele`
- ARC-Easy / ARC_Challenge, copied from `allenai/ai2_arc` + `Eurolingua/arcx`
- PolyMath, copied from `Qwen/PolyMath`


## Schema
Every multiple-choice benchmark gets the same schema, so one set of `lm-eval` templates covers all:

```output
    <data>
    id               item key shared by all languages (mmlu "subject/idx", belebele "link#qnum", ARC id)
    question         question text (the passage is separate, in `passage`, for belebele)
    choices          option texts, already in the NEW order
    answer           index of the correct option in `choices` (NEW order)
    answer_original  index of the correct option in the source order
    perm             perm[new_position] = source position, so choices[i] == source_choices[perm[i]]
    shuffled         False when the item kept its source order (see "Pinned options")

    <metadata>
    lang
    translation_tier
    source_dataset

    <custom>
    + mmlu: subject, subject_category, idx
    + belebele: passage
```

## Shuffle rules

Some source distributions are skewed: for example, MMLU-Redux-EU has D 27.6%, so "always D" would beat chance. 

- One permutation per item, decided on the English item and applied to every language, so the files stay line-parallel: the `i`-th row has the same correct letter everywhere.
- The correct option always moves to a different position than in the source.
- Its new position is chosen greedily to balance the answer letters within each split. Distractors fill the remaining positions at random.
- Pinned options: an option that refers to the others ("All of the above", "Both A and B", "Neither of these") keeps its source position; the rest are shuffled around it. If the correct answer itself is pinned it cannot move, so the item keeps its source order (shuffled=False).
- Deterministic: items are processed in an order fixed by a hash of (seed, id).
- Few-shot splits drop items with an empty option in any language (in every language); test splits are kept whole.


## Evaluation settings

We are collecting both generative- and likelikhood-based benchmarks, to cover the evaluation of the base and fine-tuned models.

PolyMath is free-form maths (no options): copied as is, one `test` split with a `level` column.