The dataset was sourced from GitHub repositories and scraped using https://github.com/cheop-byeon/RFCRationaleBuilder.
It contains two types: coding repositories and Internet-Drafts/RFCs technical documents drafting repositories.
We use the "diff" format to represent the "modification" ("snippet" as the second version, first version was trying to restore a complete expression, e.g., a whole function or a natural paragraph.)
The dataset can be downloaded using our provided script: download_CodeConvo.py. There are two directions: i2c and c2i. Here we borrow the GitHub terms, i (issue) refers to comment/review/feedback, c (code) referes to the edited code or document.
# Download entire CodeConvo dataset
python download_CodeConvo.py
huggingface-cli# Install CLI (if needed)
pip install -U "huggingface_hub[cli]"
# Download full CodeConvo repository to local folder
huggingface-cli download jiebi/CodeConvo --repo-type dataset --local-dir ./dataset/CodeConvo
Train split (structure: train/{direction}/)
# Get train/c2i folder (default direction)
python download_CodeConvo.py --split train
# Get train/i2c folder
python download_CodeConvo.py --split train --direction i2c
Dev/Test splits (structure: {repo}/{direction}/{split}/)
# Get test data for specific repo and direction
python download_CodeConvo.py --split test --repo ids --direction c2i
# Get dev data for kubernetes repo (i2c direction)
python download_CodeConvo.py --split dev --repo kubernetes --direction i2c
# Other repos: ids, ids-supp, swe, kubernetes
python download_CodeConvo.py --split test --repo swe --direction i2c
Path resolution without downloading
# Only resolve path, skip download
python download_CodeConvo.py --split train --no-download
python download_CodeConvo.py --split test --repo ids --direction c2i --no-download
--split train: --direction is optional (defaults to c2i)--split dev/test: Both --repo and --direction are requiredc2i, i2cids, ids-supp, swe, kubernetesThe data was used for retrieval tasks, for training (https://github.com/cheop-byeon/FlagEmbedding) and for evaluation (https://github.com/cheop-byeon/mteb-R2Gen).
For data, use cases, and models, please refer to:
@article{bian2025automated,
title={Automated insights into github collaboration dynamics},
author={Bian, Jie and Arefev, Nikolay and M{\"u}hlh{\"a}user, Max and Welzl, Michael},
journal={IEEE Access},
year={2025},
publisher={IEEE}
}
9 commits
The dataset was sourced from GitHub repositories and scraped using https://github.com/cheop-byeon/RFCRationaleBuilder.
It contains two types: coding repositories and Internet-Drafts/RFCs technical documents drafting repositories.
We use the "diff" format to represent the "modification" ("snippet" as the second version, first version was trying to restore a complete expression, e.g., a whole function or a natural paragraph.)
The dataset can be downloaded using our provided script: download_CodeConvo.py. There are two directions: i2c and c2i. Here we borrow the GitHub terms, i (issue) refers to comment/review/feedback, c (code) referes to the edited code or document.
# Download entire CodeConvo dataset
python download_CodeConvo.py
huggingface-cli# Install CLI (if needed)
pip install -U "huggingface_hub[cli]"
# Download full CodeConvo repository to local folder
huggingface-cli download jiebi/CodeConvo --repo-type dataset --local-dir ./dataset/CodeConvo
Train split (structure: train/{direction}/)
# Get train/c2i folder (default direction)
python download_CodeConvo.py --split train
# Get train/i2c folder
python download_CodeConvo.py --split train --direction i2c
Dev/Test splits (structure: {repo}/{direction}/{split}/)
# Get test data for specific repo and direction
python download_CodeConvo.py --split test --repo ids --direction c2i
# Get dev data for kubernetes repo (i2c direction)
python download_CodeConvo.py --split dev --repo kubernetes --direction i2c
# Other repos: ids, ids-supp, swe, kubernetes
python download_CodeConvo.py --split test --repo swe --direction i2c
Path resolution without downloading
# Only resolve path, skip download
python download_CodeConvo.py --split train --no-download
python download_CodeConvo.py --split test --repo ids --direction c2i --no-download
--split train: --direction is optional (defaults to c2i)--split dev/test: Both --repo and --direction are requiredc2i, i2cids, ids-supp, swe, kubernetesThe data was used for retrieval tasks, for training (https://github.com/cheop-byeon/FlagEmbedding) and for evaluation (https://github.com/cheop-byeon/mteb-R2Gen).
For data, use cases, and models, please refer to:
@article{bian2025automated,
title={Automated insights into github collaboration dynamics},
author={Bian, Jie and Arefev, Nikolay and M{\"u}hlh{\"a}user, Max and Welzl, Michael},
journal={IEEE Access},
year={2025},
publisher={IEEE}
}
9 commits