AI Potluck
Model components / Benchmark / eval datasets

MultiPL-E

Northeastern University / Brown / Wellesley

Translation of HumanEval's 161 problems and MBPP's 397 into 22 to 23 programming languages including Ada, C++, C#, Go, Haskell, Java, JavaScript, Lua, OCaml, PHP, R, Ruby, Rust, Scala, Shell, Swift and TypeScript. It lets code-generation performance be compared across languages rather than in Python alone, and is integrated into the BigCode evaluation harness.

The repository terms carry a clause forbidding the contents' use as machine-learning training data, which the openness axis resolves. Verified 2026-08-13 via the nuprl/MultiPL-E dataset card and the repository LICENSE.

Openness

4 medium confidence
4.0
data
downloadable(HF parquet, 47 subsets)
license
BSD-3-Clause-with-ML-Restriction(the nuprl/MultiPL-E repository LICENSE, whose clause 4 forbids using the contents as training data for any machine learning model
datasheet
present(card documents prompts/doctests/tests/stop-tokens)
redistributable
yes(HF)

Where a repository license and a distribution-point license differ, the repository license governs, and nuprl/MultiPL-E ships a LICENSE headed "BSD 3-Clause License with Machine Learning Restriction" whose clause 4 says "The contents of this repository may not be used as training data for any machine learning model, including but not limited to neural networks." That is the license recorded here; the MIT on the Hugging Face data artifact is the weaker statement of the two. The dataset openness scale has no rung for a license that permits redistribution while bounding what the corpus may be used for, so the score is a curator judgment rather than a lookup: freely downloadable and redistributable, but carrying a use restriction that stops it short of an unqualified open case.

  • https://huggingface.co/datasets/nuprl/MultiPL-E recorded 2026-08-13

    `license:mit` in the repo tags; the embedded repo state reads `"gated":false`, so the parquet downloads without a barrier; the card documents the 47 subsets and the prompt/doctests/tests/stop_tokens schema; 33,673 downloads in the trailing 30 days

  • https://github.com/nuprl/MultiPL-E/blob/main/LICENSE recorded 2026-08-13

    LICENSE headed "BSD 3-Clause License with Machine Learning Restriction"; clause 4 reads "The contents of this repository may not be used as training data for any machine learning model, including but not limited to neural networks."

Adoption

3 high confidence
3.0

The Hugging Face page reports 33,673 downloads in the trailing 30 days, which puts it in the 10K-100K band, level 3 on the dataset adoption scale.

Capability

not assessed

MultiPL-E is a dataset and translation framework rather than a system with capabilities to measure, so this axis is left unscored.

Verified 2026-08-13