Skip to content

Latest commit

 

History

History
86 lines (60 loc) · 2.53 KB

CONTRIBUTING.md

File metadata and controls

86 lines (60 loc) · 2.53 KB

Contributing to the paper list

We welcome contributions to this list!

The papers are maintained under the papers directory. The README.md file is generated from the papers in the papers directory using the python script generate_readme.py.

The structure of the papers directory is as follows:

papers
├── phenomena-of-interest
│   ├── in-context-learning
│   ├── chain-of-thought
│   ├── hallucination
│   ├── scaling-laws
│   ├── knowledge
│   ├── training-dynamics
│   ├── learning
│   ├── reversal-curse
│   └── other-phenomena
├── representational-capacity
│   ├── what-can-transformer-do
│   └── what-can-transformer-not-do
├── architectural-effectivity
│   ├── layer-normalization
│   ├── tokenization
│   └── linear-attention
├── training-paradigms
├── mechanistic-engineering
├── miscellanea
└── README.template

There is a papers.csv file in each directory that contains the metadata for each paper. The metadata includes the title, authors, date, and URL. The README.template file contains the template for the README file.

All you need to do is update the papers.csv file in the corresponding directory and pull request to the main branch. The GitHub Action will automatically generate the README.md file.

Note:

  • The papers can be overlapping across different topics. In such cases, the paper can be added to multiple directories.

  • The information in the papers.csv file should be in the following format:

Title,Date,Url,Author
  • The papers will be sorted by date automatically.

Here is a very concrete example on how to contribute:

  1. Fork the repository.

  2. Clone the forked repository to your local machine.

  3. Add a new paper to the papers/phenomena-of-interest/learning/papers.csv like this:

Title,Date,Url,Author
... # papers already in the file

Initialization is Critical to Whether Transformers Fit Composite Functions by Inference or Memorizing,2024-05-08,http://arxiv.org/abs/2405.05409,Zhongwang Zhang; Pengxiao Lin; Zhiwei Wang; Yaoyu Zhang; Zhi-Qin John Xu
  1. Pull request to the main branch.

You can run the generation script locally to check if the README file is generated correctly:

python generate_readme.py

or dry-run the script locally:

python generate_readme.py --dry-run

If you have any questions, feel free to open an issue or reach out to the maintainers.

Thank you for contributing!