Dataset Card for FEA-Bench

A Benchmark for Evaluating Repository-Level Code Generation for Feature Implementation.

Dataset Details

Dataset Description

The FEA-Bench is a benchmark with a test set that contains 1,401 task instances from 83 Github repositories. This benchmark aims to evaluate the capabilities of repository-level incremental code development. The task instances are collected from Github pull requests, which have the purpose of new feature implementation. Each task instance includes the repo and the base commit sha256, and the PR number and the status of unit test.

• Curated by: the authors of the FEA-Bench paper: Wei Li, Xin Zhang, Zhongxin Guo, Shaoguang Mao and their collaborators.
• Language(s) (NLP): English
• License: Others; We list all licenses of involved github repositories in the last part.

Uses

This dataset is designed to evaluate performances of LLMs on repository-level code development, which is a complicated software engineering task.

• Repository-level incremental code development: The FEA-Bench can be used to evaluate a model for the the capabilities of repository-level incremental code development. Success on this task is typically measured by achieving a high/low resolved ratio. The leaderboard will soon be published as a website.

Direct Use

Use scripts from FEA-Bench repo to get info for task instances and organize them into prompt, which can be used to LLMs' inference. Also, you can get info or use agents to directly solve the PRs with code changes.

Out-of-Scope Use

This dataset is not aimed at training for LLMs. You should not take the FEA-Bench as the training dataset to avoid contamination.

Dataset Structure

An example:

{
    "instance_id": "huggingface__accelerate-270",
    "pull_number": 270,
    "repo": "huggingface/accelerate",
    "version": null,
    "base_commit": "515fcca9ed2b36c274c595dbdff75f1c2da635de",
    "environment_setup_commit": "08101b9dde2b1a9658c2e363e3e9f5663ba06073",
    "FAIL_TO_PASS": [
        "tests/test_state_checkpointing.py::CheckpointTest::test_can_resume_training",
        "tests/test_state_checkpointing.py::CheckpointTest::test_invalid_registration",
        "tests/test_state_checkpointing.py::CheckpointTest::test_with_scheduler"
    ],
    "PASS_TO_PASS": []
}

Dataset Creation

Curation Rationale

Implementing new features in repository-level codebases is a crucial application of code generation models. However, current benchmarks lack a dedicated evaluation framework for this capability. To fill this gap, we introduce FEA-Bench, a benchmark designed to assess the ability of large language models (LLMs) to perform incremental development within code repositories.

Source Data

Data Collection and Processing

We collect pull requests from 83 GitHub repositories and use rule-based and intent-based filtering to construct task instances focused on new feature development. Each task instance containing code changes is paired with relevant unit test files to ensure that the solution can be verified.

Who are the source data producers?

Authors of 83 Github repositories list in the last part.

Personal and Sensitive Information

The dataset does not include any personal or sensitive information.

Bias, Risks, and Limitations

• The quantity of high-quality data suitable for repository-level incremental development is limited. High-quality and usable pull requests for new feature development are relatively scarce. Many repository-level code developments for implementing new functionalities were committed during the early stages of repositories, without going through the rigorous code review process typical of the open-source community, resulting in lower data quality that cannot be utilized.
• Furthermore, the software's early-stage developments might not even have been conducted using the GitHub platform, posing a challenge for data collection and utilization.
• The repository-level incremental code development may not just include new feature implementation tasks.
• Only Python repositories are involved in FEA-Bench.
• The inference results of the task instances from the benchmark may contain code that is harmful to computer systems.

Recommendations

Evaluation by docker is recommended, just like SWE-bench. We will also publish a patch for SWE-bench to make it compatible for our tasks' evaluation.

BibTeX:

To be appeared after publishing the ArXiv paper.

APA:

To be appeared after publishing the ArXiv paper.

Dataset Card Contact

For further information or questions, please contact Xin Zhang (xinzhang3@microsoft.com).

All involved Github repositories in the FEA-Bench

Repo Name	License	Topic
astropy/astropy	BSD-3-Clause	Scientific/Engineering::Astronomy
django/django	BSD-3-Clause	Internet::WWW/HTTP
matplotlib/matplotlib	Other	Scientific/Engineering::Visualization
mwaskom/seaborn	BSD-3-Clause	Scientific/Engineering::Visualization
pallets/flask	BSD-3-Clause	Internet::WWW/HTTP
pvlib/pvlib-python	BSD-3-Clause	Scientific/Engineering::Physics
pydata/xarray	Apache-2.0	Scientific/Engineering::Information Analysis
pydicom/pydicom	Others	Scientific/Engineering::Medical Science Apps.
pylint-dev/astroid	LGPL-2.1	Software Development::Libraries
pylint-dev/pylint	GPL-2.0	Software Development::Quality Assurance
pyvista/pyvista	MIT	Scientific/Engineering::Information Analysis
scikit-learn/scikit-learn	BSD-3-Clause	Scientific/Engineering::Artificial Intelligence
sphinx-doc/sphinx	BSD-2-Clause	Text Processing::Markup
sqlfluff/sqlfluff	MIT	Software Development::Quality Assurance
sympy/sympy	Others	Scientific/Engineering::Mathematics
Aider-AI/aider	Apache-2.0	Software Development::Code Generators
Cog-Creators/Red-DiscordBot	GPL-3.0	Communications::Chat
DLR-RM/stable-baselines3	MIT	Scientific/Engineering::Artificial Intelligence
EleutherAI/lm-evaluation-harness	MIT	Scientific/Engineering::Artificial Intelligence
Project-MONAI/MONAI	Apache-2.0	Scientific/Engineering::Medical Science Apps.
PyThaiNLP/pythainlp	Apache-2.0	Text Processing::Linguistic
RDFLib/rdflib	BSD-3-Clause	Software Development::Libraries
Textualize/rich	MIT	Software Development::Libraries
Textualize/textual	MIT	Software Development::User Interfaces
TileDB-Inc/TileDB-Py	MIT	Software Development::Libraries
astronomer/astronomer-cosmos	Apache-2.0	Software Development::Build Tools
atlassian-api/atlassian-python-api	Apache-2.0	Internet::WWW/HTTP
aws-cloudformation/cfn-lint	MIT-0	Software Development::Quality Assurance
aws-powertools/powertools-lambda-python	MIT-0	Software Development::Libraries
aws/sagemaker-python-sdk	Apache-2.0	Scientific/Engineering::Artificial Intelligence
biopragmatics/bioregistry	MIT	Scientific/Engineering::Bio-Informatics
boto/boto3	Apache-2.0	Software Development::Libraries
boto/botocore	Apache-2.0	Software Development::Libraries
cocotb/cocotb	BSD-3-Clause	Scientific/Engineering::Electronic Design Automation (EDA)
conan-io/conan	MIT	Software Development::Build Tools
deepset-ai/haystack	Apache-2.0	Scientific/Engineering::Artificial Intelligence
docker/docker-py	Apache-2.0	Software Development::Libraries
dpkp/kafka-python	Apache-2.0	Software Development::Libraries
embeddings-benchmark/mteb	Apache-2.0	Scientific/Engineering::Artificial Intelligence
facebookresearch/hydra	MIT	Software Development::Libraries
fairlearn/fairlearn	MIT	Scientific/Engineering::Artificial Intelligence
falconry/falcon	Apache-2.0	Internet::WWW/HTTP
google-deepmind/optax	Apache-2.0	Scientific/Engineering::Artificial Intelligence
googleapis/python-aiplatform	Apache-2.0	Scientific/Engineering::Artificial Intelligence
googleapis/python-bigquery	Apache-2.0	Internet::WWW/HTTP
gradio-app/gradio	Apache-2.0	Scientific/Engineering::Human Machine Interfaces
graphql-python/graphene	MIT	Software Development::Libraries
huggingface/accelerate	Apache-2.0	Scientific/Engineering::Artificial Intelligence
huggingface/datasets	Apache-2.0	Scientific/Engineering::Artificial Intelligence
huggingface/huggingface_hub	Apache-2.0	Scientific/Engineering::Artificial Intelligence
huggingface/pytorch-image-models	Apache-2.0	Software Development::Libraries
huggingface/trl	Apache-2.0	Scientific/Engineering::Artificial Intelligence
joblib/joblib	BSD-3-Clause	Software Development::Libraries
joke2k/faker	MIT	Software Development::Testing
lark-parser/lark	MIT	Text Processing::Linguistic
minio/minio-py	Apache-2.0	Software Development::Libraries
open-mmlab/mmengine	Apache-2.0	Utilities
openvinotoolkit/datumaro	MIT	Scientific/Engineering::Image Processing
pgmpy/pgmpy	MIT	Scientific/Engineering::Artificial Intelligence
pre-commit/pre-commit	MIT	Software Development::Quality Assurance
prometheus/client_python	Apache-2.0	System::Monitoring
prompt-toolkit/python-prompt-toolkit	BSD-3-Clause	Software Development::User Interfaces
pygments/pygments	BSD-2-Clause	Software Development::Documentation
pyocd/pyOCD	Apache-2.0	Software Development::Debuggers
pypa/hatch	MIT	Software Development::Build Tools
pyro-ppl/pyro	Apache-2.0	Scientific/Engineering::Artificial Intelligence
python-hyper/h2	MIT	Internet::WWW/HTTP
roboflow/supervision	MIT	Scientific/Engineering::Image Processing
rytilahti/python-miio	GPL-3.0	Home Automation
saleweaver/python-amazon-sp-api	MIT	Internet::WWW/HTTP
scrapy/scrapy	BSD-3-Clause	Software Development::Libraries
scverse/scanpy	BSD-3-Clause	Scientific/Engineering::Bio-Informatics
slackapi/bolt-python	MIT	Communications::Chat
slackapi/python-slack-sdk	MIT	Communications::Chat
snowflakedb/snowflake-connector-python	Apache-2.0	Software Development::Libraries
softlayer/softlayer-python	MIT	Software Development::Libraries
spec-first/connexion	Apache-2.0	Internet::WWW/HTTP
statsmodels/statsmodels	BSD-3-Clause	Scientific/Engineering::Information Analysis
tfranzel/drf-spectacular	BSD-3-Clause	Software Development::Documentation
tobymao/sqlglot	MIT	Database::Database Engines/Servers
tornadoweb/tornado	Apache-2.0	Internet::WWW/HTTP
tortoise/tortoise-orm	Apache-2.0	Database::Front-Ends
wagtail/wagtail	BSD-3-Clause	Internet::WWW/HTTP