tldr / README.md
shuyanzh's picture
Create README.md
62e85e8
metadata
annotations_creators: []
language_creators:
  - crowdsourced
  - expert-generated
language:
  - code
license:
  - mit
multilinguality:
  - monolingual
size_categories:
  - unknown
source_datasets:
  - original
task_categories:
  - text2text-generation
task_ids: []
pretty_name: DocPrompting-CoNaLa
tags:
  - code-generation
  - doc retrieval
  - retrieval augmented generation

Dataset Description

Dataset Summary

This is the natural language to bash generation dataset we harvested from the English subset of tldr We split the dataset by bash commands. Every command in the dev and test set is held out from the training set.

Supported Tasks and Leaderboards

This dataset is used to evaluate code generations.

Languages

English - Bash

Dataset Structure

dataset = load_dataset("neulab/tldr")
DatasetDict({
    train: Dataset({
        features: ['question_id', 'nl', 'cmd', 'oracle_man', 'cmd_name', 'tldr_cmd_name', 'manual_exist', 'matching_info'],
        num_rows: 6414
    })
    test: Dataset({
        features: ['question_id', 'nl', 'cmd', 'oracle_man', 'cmd_name', 'tldr_cmd_name', 'manual_exist', 'matching_info'],
        num_rows: 928
    })
    validation: Dataset({
        features: ['question_id', 'nl', 'cmd', 'oracle_man', 'cmd_name', 'tldr_cmd_name', 'manual_exist', 'matching_info'],
        num_rows: 1845
    })
})

code_docs = load_dataset("neulab/docprompting-conala", "docs")
DatasetDict({
    train: Dataset({
        features: ['doc_id', 'doc_content'],
        num_rows: 439064
    })
})

Data Fields

train/dev/test:

  • nl: The natural language intent
  • cmd: The reference code snippet
  • question_id: the unique id of a question
  • oracle_man: The doc_id of the functions used in the reference code snippet. The corresponding contents are in doc split
  • cmd_name: the bash command of this code snippet
  • tldr_cmd_name: the bash command used in tldr github repo. The cmd_name and tldr_cmd_name can be different due to naming difference
  • manual_exist: whether the manual exists in https://manned.org
  • matching_info: each code snippets have multiple tokens, this is the detailed reference doc matching on each token.

docs:

  • doc_id: the id of a doc
  • doc_content: the content of the doc

Dataset Creation

The dataset was curated from tldr. The project aims to provide frequent usage of bash commands with natural language intents. For more details, please check the repo.

Citation Information

@article{zhou2022doccoder,
  title={DocCoder: Generating Code by Retrieving and Reading Docs},
  author={Zhou, Shuyan and Alon, Uri and Xu, Frank F and Jiang, Zhengbao and Neubig, Graham},
  journal={arXiv preprint arXiv:2207.05987},
  year={2022}
}