Younger

This repository maintains all code related to using the Younger dataset and its related applications.

Go to the website "Datasets - Yangs AI" for Detailed Documentation of Younger's Applications.

Due to ongoing updates to this branch, the code may not match the initial version presented in the paper. Therefore, if you need the version that corresponds to what is mentioned in the paper, please switch to the paper branch for reference.

Installation

Option A: Install from PyPI (recommended for users)

Install the core package:

pip install younger

Install optional feature groups as needed (these pull sub-packages published to PyPI):

# Logic toolchains
pip install "younger[logics]"

# Deep-learning apps
pip install "younger[apps]"

Core and Bench publication is deferred. Extras for those packages will be added after their releases; younger[logics] currently installs IR only.

Option B: Install from source (recommended for developers)

This repo includes multiple sub-packages. Use editable installs for local development.

git clone --recursive git@github.com:Yangs-AI/Younger.git
cd Younger
git checkout dev
git submodule update --init --recursive

# kernel
pip install -e .

# apps: younger_apps_dl
pip install -e younger/apps/dl

# logics: younger_logics_core + younger_logics_ir
pip install -e younger/logics/core
pip install -e younger/logics/ir

Publishing Strategy (for maintainers)

Younger is maintained in independent repositories connected by Git submodules. Each package owns its version and publishes independently to GitHub and PyPI:

• younger (kernel)
• younger-logics-core (publication deferred)
• younger-logics-ir
• younger-apps-dl
• younger-tools-bench (planned)

The current release scope is younger, younger-logics-ir, and younger-apps-dl at 0.1.0; subsequent versions do not need to match. Internal dependency ranges declare compatibility. When new dependency versions are needed, publish the kernel first, then logics, then apps/tools. Kernel extras install compatible components. releases/current.txt records the candidate combination checked by CI; named files in releases/ preserve specific combinations. See RELEASE.md for CI validation, Trusted Publishing setup, release tags, and the coordinated release procedure. Conda distribution is deferred.

Dependency Map

• younger: kernel, with optional extras for components
• younger-logics-core and younger-logics-ir: depend on younger
• younger-apps-dl and younger-tools-bench: depend on younger and younger-logics-ir
• younger-apps-dl will also depend on younger-logics-core when that integration is implemented and a compatible Core release is available.

Runtime Dependencies (Summary)

• younger: tqdm, click, fsspec, psutil, tomlkit, requests, python-semantic-release
• younger-logics-core: tqdm, click, psutil, pandas, networkx
• younger-logics-ir: onnx, tqdm, click, networkx
• younger-apps-dl: torch, pandas, scikit-learn, networkx, torch-geometric, tabulate, pydantic, younger-logics-ir
• younger-tools-bench: tqdm, click, psutil, pandas, networkx, younger-logics-ir

Development & Contributing

Please see CONTRIBUTING.md for development setup, branching, testing, and contribution workflow.

Release

Maintainers can follow RELEASE.md for publishing order and release checklist.

Datasets

Browse and download the datasets at here. It may take some time to prepare the datasets for the first release. Please stay tuned. In future releases, we will provide features to automatically download and preprocess the datasets.

Citation

If you need to use our code or dataset, please cite the following paper:

1. Yang, Z., Gao, W., Peng, L., Huang, Y., Tang, F., & Zhan, J. (2024). Younger: The First Dataset for Artificial Intelligence-Generated Neural Network Architecture. arXiv

@misc{yang2024youngerdatasetartificialintelligencegenerated,
      title={Younger: The First Dataset for Artificial Intelligence-Generated Neural Network Architecture}, 
      author={Zhengxin Yang and Wanling Gao and Luzhou Peng and Yunyou Huang and Fei Tang and Jianfeng Zhan},
      year={2024},
      eprint={2406.15132},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2406.15132}, 
}