README.md

metadata

license: mit
pretty_name: HumanEvalPack
language_creators:
  - expert-generated
multilinguality:
  - multilingual
language:
  - code
tags:
  - code

Dataset Card for HumanEvalPack

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
• Additional Information
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Repository: https://github.com/bigcode-project/octopack
• Paper: OctoPack: Instruction Tuning Code Large Language Models
• Point of Contact: Niklas Muennighoff

Dataset Summary

HumanEvalPack is an extension of OpenAI's HumanEval to cover 6 total languages across 3 tasks. The Python split is exactly the same as OpenAI's Python HumanEval. The other splits are translated by humans (similar to HumanEval-X but with additional cleaning, see here). Refer to the OctoPack paper for more details.

• Languages: Python, JavaScript, Java, Go, C++, Rust
• OctoPack🐙🎒:

Data	CommitPack	4TB of GitHub commits across 350 programming languages
	CommitPackFT	Filtered version of CommitPack for high-quality commit messages that resemble instructions
Model	OctoCoder	StarCoder (16B parameters) instruction tuned on CommitPackFT + OASST
	OctoGeeX	CodeGeeX2 (6B parameters) instruction tuned on CommitPackFT + OASST
Evaluation	HumanEvalPack	Extension of OpenAI's HumanEval to cover 3 scenarios across 6 languages

Usage

# pip install -q datasets
from datasets import load_dataset
# Languages: "python", "js", "java", "go", "cpp", "rust"
ds = load_dataset("bigcode/humanevalpack", "python")["test"]
ds[0]

Dataset Structure

Data Instances

An example looks as follows:

{
  "task_id": "Python/0",
  "prompt": "from typing import List\n\n\ndef has_close_elements(numbers: List[float], threshold: float) -> bool:\n    \"\"\" Check if in given list of numbers, are any two numbers closer to each other than\n    given threshold.\n    >>> has_close_elements([1.0, 2.0, 3.0], 0.5)\n    False\n    >>> has_close_elements([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3)\n    True\n    \"\"\"\n",
  "declaration": "from typing import List\n\n\ndef has_close_elements(numbers: List[float], threshold: float) -> bool:\n",
  "canonical_solution": "    for idx, elem in enumerate(numbers):\n        for idx2, elem2 in enumerate(numbers):\n            if idx != idx2:\n                distance = abs(elem - elem2)\n                if distance < threshold:\n                    return True\n\n    return False\n",
  "buggy_solution": "    for idx, elem in enumerate(numbers):\n        for idx2, elem2 in enumerate(numbers):\n            if idx != idx2:\n                distance = elem - elem2\n                if distance < threshold:\n                    return True\n\n    return False\n",
  "bug_type": "missing logic",
  "failure_symptoms": "incorrect output",
  "entry_point": "has_close_elements",
  "import": ""
  "test_setup": ""
  "test": "\n\n\n\n\ndef check(has_close_elements):\n    assert has_close_elements([1.0, 2.0, 3.9, 4.0, 5.0, 2.2], 0.3) == True\n    assert has_close_elements([1.0, 2.0, 3.9, 4.0, 5.0, 2.2], 0.05) == False\n    assert has_close_elements([1.0, 2.0, 5.9, 4.0, 5.0], 0.95) == True\n    assert has_close_elements([1.0, 2.0, 5.9, 4.0, 5.0], 0.8) == False\n    assert has_close_elements([1.0, 2.0, 3.0, 4.0, 5.0, 2.0], 0.1) == True\n    assert has_close_elements([1.1, 2.2, 3.1, 4.1, 5.1], 1.0) == True\n    assert has_close_elements([1.1, 2.2, 3.1, 4.1, 5.1], 0.5) == False\n\ncheck(has_close_elements)",
  "example_test": "def check(has_close_elements):\n    assert has_close_elements([1.0, 2.0, 3.0], 0.5) == False\n    assert has_close_elements([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3) == True\ncheck(has_close_elements)\n",
  "signature": "has_close_elements(numbers: List[float], threshold: float) -> bool",
  "docstring": "Check if in given list of numbers, are any two numbers closer to each other than\ngiven threshold.\n>>> has_close_elements([1.0, 2.0, 3.0], 0.5)\nFalse\n>>> has_close_elements([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3)\nTrue",
  "instruction": "Write a Python function `has_close_elements(numbers: List[float], threshold: float) -> bool` to solve the following problem:\nCheck if in given list of numbers, are any two numbers closer to each other than\ngiven threshold.\n>>> has_close_elements([1.0, 2.0, 3.0], 0.5)\nFalse\n>>> has_close_elements([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3)\nTrue"
}

Data Fields

The data fields are the same among all splits:

• task_id: Indicates the language (Python/JavaScript/Java/Go/C++/Rust) and task id (from 0 to 163) of the problem
• prompt: the prompt for models relying on code continuation
• declaration: the declaration of the function (same as prompt but without the docstring)
• canonical_solution: the correct solution passing all unit tests for the problem
• buggy_solution: same as canonical_solution but with a subtle human-written bug causing the unit tests to fail
• bug_type: the type of the bug in buggy_solution (one of [missing logic, excess logic, value misuse, operator misuse, variable misuse, function misuse])
• failure_symptoms: the problem the bug causes (one of [incorrect output, stackoverflow, infinite loop])
• entry_point: the name of the function
• import: imports necessary for the solution (only present for Go)
• test_setup: imports necessary for the test execution (only present for Go)
• test: the unit tests for the problem
• example_test: additional unit tests different from test that could be e.g. provided to the model (these are not used in the paper)
• signature: the signature of the function
• docstring: the docstring describing the problem
• instruction: an instruction for HumanEvalSynthesize in the form Write a {language_name} function {signature} to solve the following problem:\n{docstring}

Citation Information

@article{muennighoff2023octopack,
      title={OctoPack: Instruction Tuning Code Large Language Models}, 
      author={Niklas Muennighoff and Qian Liu and Armel Zebaze and Qinkai Zheng and Binyuan Hui and Terry Yue Zhuo and Swayam Singh and Xiangru Tang and Leandro von Werra and Shayne Longpre},
      journal={arXiv preprint arXiv:2308.07124},
      year={2023}
}