Skip to content


Thanks for considering contributing! We want AllenNLP to be the way to do cutting-edge NLP research, but we cannot get there without community support.

How Can I Contribute?

Bug fixes and new features

Did you find a bug?

First, do a quick search to see whether your issue has already been reported. If your issue has already been reported, please comment on the existing issue.

Otherwise, open a new GitHub issue. Be sure to include a clear title and description. The description should include as much relevant information as possible. The description should explain how to reproduce the erroneous behavior as well as the behavior you expect to see. Ideally you would include a code sample or an executable test case demonstrating the expected behavior.

Do you have a suggestion for an enhancement?

We use GitHub issues to track enhancement requests. Before you create an enhancement request:

  • Make sure you have a clear idea of the enhancement you would like. If you have a vague idea, consider discussing it first on a GitHub issue.

  • Check the documentation to make sure your feature does not already exist.

  • Do a quick search to see whether your enhancement has already been suggested.

When creating your enhancement request, please:

  • Provide a clear title and description.

  • Explain why the enhancement would be useful. It may be helpful to highlight the feature in other libraries.

  • Include code examples to demonstrate how the enhancement would be used.

Making a pull request

When you're ready to contribute code to address an open issue, please follow these guidelines to help us be able to review your pull request (PR) quickly.

  1. Initial setup (only do this once)

    Expand details 👇

    If you haven't already done so, please fork this repository on GitHub.

    Then clone your fork locally with

    git clone


    git clone

    At this point the local clone of your fork only knows that it came from your repo,, but doesn't know anything the main repo, You can see this by running

    git remote -v

    which will output something like this:

    origin (fetch)
    origin (push)

    This means that your local clone can only track changes from your fork, but not from the main repo, and so you won't be able to keep your fork up-to-date with the main repo over time. Therefor you'll need to add another "remote" to your clone that points to To do this, run the following:

    git remote add upstream

    Now if you do git remote -v again, you'll see

    origin (fetch)
    origin (push)
    upstream (fetch)
    upstream (push)

    Finally, you'll need to create a Python 3 virtual environment suitable for working on AllenNLP. There a number of tools out there that making working with virtual environments easier, but the most direct way is with the venv module in the standard library.

    Once your virtual environment is activated, you can install your local clone in "editable mode" with

    pip install -e . 
    pip install -r dev-requirements.txt

    The "editable mode" comes from the -e argument to pip, and essential just creates a symbolic link from the site-packages directory of your virtual environment to the source code in your local clone. That way any changes you make will be immediately reflected in your virtual environment.

  2. Ensure your fork is up-to-date

    Expand details 👇

    Once you've added an "upstream" remote pointing to, keeping your fork up-to-date is easy:

    git checkout main  # if not already on main
    git pull --rebase upstream main
    git push
  3. Create a new branch to work on your fix or enhancement

    Expand details 👇

    Commiting directly to the main branch of your fork is not recommended. It will be easier to keep your fork clean if you work on a seperate branch for each contribution you intend to make.

    You can create a new branch with

    # replace BRANCH with whatever name you want to give it
    git checkout -b BRANCH
    git push -u origin BRANCH
  4. Test your changes

    Expand details 👇

    Our continuous integration (CI) testing runs a number of checks for each pull request on GitHub Actions. You can run most of these tests locally, which is something you should do before opening a PR to help speed up the review process and make it easier for us.

    First, you should run black to make sure you code is formatted consistently. Many IDEs support code formatters as plugins, so you may be able to setup black to run automatically everytime you save. black.vim will give you this functionality in Vim, for example. But black is also easy to run directly from the command line. Just run this from the root of your clone:

    black .

    Our CI also uses flake8 to lint the code base and mypy for type-checking. You should run both of these next with

    make lint


    make typecheck

    We also strive to maintain high test coverage, so most contributions should include additions to the unit tests. These tests are run with pytest, which you can use to locally run any test modules that you've added or changed.

    For example, if you've fixed a bug in allennlp/nn/, you can run the tests specific to that module with

    pytest -v tests/nn/

    Our CI will automatically check that test coverage stays above a certain threshold (around 90%). To check the coverage locally in this example, you could run

    pytest -v --cov allennlp.nn.util tests/nn/

    If your contribution involves changes to any docstrings, you should make sure the API documentation can build without errors. For that, just run

    make build-docs

    If the build fails, it's most likely due to small formatting issues. If the error message isn't clear, feel free to comment on this in your pull request.

    You can also serve and view the docs locally with

    make serve-docs

    And finally, please update the CHANGELOG with notes on your contribution in the "Unreleased" section at the top.

    After all of the above checks have passed, you can now open a new GitHub pull request. Make sure you have a clear description of the problem and the solution, and include a link to relevant issues.

    We look forward to reviewing your PR!

New models

Do you have a new state-of-the-art model?

We are always looking for new models to add to our collection. The most popular models are usually added to the official AllenNLP Models repository, and in some cases to the AllenNLP Demo.

If you think your model should be part of AllenNLP Models, please create a pull request in the models repo that includes:

  • Any code changes needed to support your new model.

  • A link to the model itself. Please do not check your model into the GitHub repository, but instead upload it in the PR conversation or provide a link to it at an external location.

In the description of your PR, please clearly explain the task your model performs along with the relevant metrics on an established dataset.