metadata

annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - code
license:
  - mit
multilinguality:
  - monolingual
source_datasets:
  - original
task_categories:
  - text2text-generation
task_ids: []
pretty_name: OpenAI HumanEval-Infilling
tags:
  - code-generation

HumanEval-Infilling

Dataset Description

Repository: https://github.com/openai/human-eval-infilling
Paper: https://arxiv.org/pdf/2207.14255

Dataset Summary

HumanEval-Infilling is a benchmark for infilling tasks, derived from HumanEval benchmark for the evaluation of code generation models.

Dataset Structure

To load the dataset you need to specify a subset. By default HumanEval-SingleLineInfilling is loaded.

from datasets import load_dataset
ds = load_dataset("humaneval_infilling", "HumanEval-RandomSpanInfilling")

DatasetDict({
    test: Dataset({
        features: ['task_id', 'entry_point', 'prompt', 'suffix', 'canonical_solution', 'test'],
        num_rows: 1640
    })
})

Subsets

This dataset has 4 subsets: HumanEval-MultiLineInfilling, HumanEval-SingleLineInfilling, HumanEval-RandomSpanInfilling, HumanEval-RandomSpanInfillingLight. The single-line, multi-line, random span infilling and its light version have 1033, 5815, 1640 and 164 tasks, respectively.

Citation

@article{bavarian2022efficient,
  title={Efficient Training of Language Models to Fill in the Middle},
  author={Bavarian, Mohammad and Jun, Heewoo and Tezak, Nikolas and Schulman, John and McLeavey, Christine and Tworek, Jerry and Chen, Mark},
  journal={arXiv preprint arXiv:2207.14255},
  year={2022}
}