More than 300 awesome developers have contributed to the gradio
library, and we'd be thrilled if you would like to be the next contributor!
Prerequisites:
- Python 3.10+
- Node.js v16.14+ (only needed if you are making changes to the frontend)
- pnpm 8.1+ (only needed if you are making changes to the frontend)
Steps to Contribute:
Generally speaking, contributing to Gradio involves four steps:
- Identify a good issue to contribute to (such as any of the issues tagged with "good first issue")
- Setup Gradio locally
- Understand the structure of the codebase & make the changes to the codebase locally
- Open a pull request (PR) to upstream your changes to the Gradio repository
Note: We welcome meaningful contributions that solve real issues and improve the codebase. Please avoid opening PRs with content generated primarily by AI language models. All contributions should demonstrate clear understanding of the problem being solved and be consistent with the relevant sections of the Gradio codebase.
You can watch this short video walkthrough of how to contribute, or keep reading below:
There are a few ways to install and run Gradio.
- Clone this repo
- Navigate to the repo directory and run:
bash scripts/install_gradio.sh
- Run the frontend (only required if you are making changes to the frontend and would like to preview them)
bash scripts/run_frontend.sh
- Install test requirements (only required if you want to run tests locally)
(Note that it is highly recommended to use a virtual environment running Python 3.10 since the versions of Gradio's dependencies are pinned)
bash scripts/install_test_requirements.sh
If you have a different Python version and conflicting packages during the installation, please first run:
bash scripts/create_test_requirements.sh
Instead of installing Gradio locally, you can alternatively use dev containers. This is supported on all platforms (macOS/Windows/Linux), as well as on GitHub Codespaces.
Prerequisites:
- An editor which supports dev containers, like VS Code
- Docker support on the host computer:
- macOS: Docker Desktop 2.0+
- Windows: Docker Desktop 2.0+
- Linux: Docker CE/EE 18.06+ and Docker Compose 1.21+
- If using VS Code, the Dev Containers extension
Steps:
- Clone repository
- Open it in your editor
- For VS Code, execute
Dev Containers: Reopen in container
command
For detailed instructions, please see the Dev Containers tutorial.
If you're a newcomer to Gradio, we recommend getting familiar with the overall structure of the repository so that you can focus on the part of the source code you'd like to contribute to.
/gradio
: contains the Python source code for the library/gradio/interface.py
: contains the Python source code for the coreInterface
class/gradio/blocks.py
: contains the Python source code for the coreBlocks
class/gradio/components/
: the directory that contains the Python source code for all of the Gradio components.
/test
: contains Python unit tests for the library/js
: contains the HTML/JS/CSS source code for the library, including the fronted code for each component in a separate directory/js/_website
: contains the code for the Gradio website (www.gradio.app). See the README in the/js/_website
folder for more details
/guides
: the written guides and tutorials that are found on Gradio's website.
You can get started by creating an app.py
file in the root:
import gradio as gr
with gr.Blocks() as demo:
gr.Button()
if __name__ == "__main__":
demo.launch()
then run:
gradio app.py
This will start the backend server in reload mode, which will watch for changes in the gradio
folder and reload the app if changes are made. By default, Gradio will launch on port 7860. You can also just use python app.py
, but this won't automatically trigger updates.
Note: if you have gradio
installed elsewhere in your system, you may need to uninstall it or at least make sure your PYTHONPATH
includes the directory where the Gradio repository is cloned, e.g.,
export PYTHONPATH="./"
If you're making frontend changes, start the frontend server:
bash scripts/run_frontend.sh
This will open a separate browser tab. By default, Gradio will launch this on port 9876. Any changes to the frontend will also reload automatically in the browser. For more information about developing in the frontend, you can refer to js/README.md.
We also have demos of all our components in the /gradio/demo
directory. To get our simple gradio Chatbot running locally:
gradio demo/chatbot_simple/run.py
We use Pytest, Playwright and Vitest to test our code.
- The Python tests are located in
/test
. To run these tests:
bash scripts/run_backend_tests.sh
- The frontend unit tests are any defined with the filename
*.test.ts
. To run them:
pnpm test
- Browser tests are located in
js/spa/test
and are defined as*spec.ts
files. To run browser tests:
pnpm test:browser
To build the frontend code before running browser tests:
pnpm test:browser:full
You can also run browser tests in the UI mode by adding the --ui
flag:
pnpm test:browser --ui
If you have made any significant visual changes to a component, we encourage you to add a new Storybook story or amend an existing one to reflect them. You can create a new story with a *.stories.svelte
file. You can run the storybook locally:
pnpm storybook
We also welcome any contributions to our website and docs.
All of the website code lives in the js/_website/
directory.
To start the website on dev mode simply cd into this directory and run:
pnpm i
pnpm dev
This will serve the website on http://localhost:5173/
(or the next available port).
When you're done with changes and want to build the website you can run:
pnpm build && pnpm preview
This will serve the website on http://localhost:4173/
(or the next available port).
Gradio's API reference is built from templates written in mdsvex. You can find all the templates in this directory:
js/_website/src/lib/templates/gradio
The templates directory is structured as follows:
├── gradio/
│ ├── 01_building-demos/
│ │ ├── 01_interface.svx
│ │ ├── 02_chatinterface.svx
│ │ ├── 03_tabbedinterface.svx
│ │ ├── 04_blocks.svx
│ ├── 02_blocks-layout/
│ ├── 03_components/
│ ├── 04_helpers/
│ ├── 05_modals/
│ ├── 06_routes/
│ ├── other/
This structure defines the pages' ordering. You can use a numeral prefix (XX_) before a name to dictate where a page is listed, but it's otherwise ignored in the url route. Note that the folder names (01_building-demos, etc) are only used for the navbar and are not in the url.
The mdsvex files use a combination of markdown and svelte. They also pull documentation directly from the source code. Adding a @document()
wrapper around any class or function in the source code will make its docstrings available in the templates.
Here's an example: the template for Image docs is here. You can see the initialization section references obj.parameters
. So to edit the description of a parameter you'll have to edit the docstring in the source code. But the page also includes a section titled 'GIF and SVG Image Formats' which is written in plain markdown and can be edited directly on the template.
If you are making changes to docstrings and want to see them on the website you have to make sure you're on an editable install of the gradio library. Just run this command from root:
pip install -e .
And then from the website directory:
pnpm dev
Guides like Quickstart are built from this directory: /guides
. The directory follows the same structure as the API reference templates, with nested folders and numerical prefixes for ordering, but the files are standard markdown files. After adding a new guide, or editing an existing one, to see the changes on the website make sure you are on an editable install of the gradio library. Run this command from root:
pip install -e .
and then from the website directory:
pnpm dev
The website supports documentation for both the latest released version on pypi as well as the main build on github. You can switch between them on the website by using the toggle on any page or by prefixing '/main' before the route in the url. For example: https://www.gradio.app/main/guides/quickstart
If you're making changes to documentation and are wondering why they're not showing up, make sure you're looking at the 'main' version of the page. Since they haven't been included in a release yet, they will only be visible there.
Gradio-Lite is a Pyodide-based library that lets you run Gradio serverless (in other words, directly in your browser).
You can start the development server by running:
bash scripts/run_lite.sh
If you make changes to the Python code during development, you will need to rebuild the Python packages loaded to Graio-Lite. To do this, run:
pnpm --filter @gradio/lite pybuild
To generate the release build, run:
bash scripts/build_lite.sh
The release build will be located in the dist
directory in the js/lite
project.
To test it, you can run a local server in the js/lite
directory:
python -m http.server --directory js/lite
and navigate to http://localhost:8000
in your browser. The demo page index.html
located in the js/lite
directory will be loaded.
All PRs should be submitted against main
, and ideally should address an open issue, unless the change is small. Direct commits to main are blocked, and PRs require an approving review to merge into main. By convention, the Gradio maintainers will review PRs when:
- An initial review has been requested
- A clear, descriptive title has been assigned to the PR
- A maintainer (@abidlabs, @aliabid94, @aliabd, @AK391, @dawoodkhan82, @pngwn, @freddyaboulton, @hannahblair, @hysts, @whitphx) is tagged in the PR comments and asked to complete a review
🧹 We ask that you make sure initial CI checks are passing before requesting a review. One of the Gradio maintainers will merge the PR when all the checks are passing. You can safely ignore the Vercel and Spaces checks, which only run under maintainers' pull requests.
Don't forget to format your code before pushing:
bash scripts/format_backend.sh
And if you made changes to the frontend:
bash scripts/format_frontend.sh
Thank you for taking the time to contribute to Gradio!
- Browse issues with the "good first issue" label. These are issues we think are good for newcomers.
- Ask the Gradio community in our Discord
- Raise an issue for a feature or a bug you want to tackle
ERROR: Error loading ASGI app. Could not import module "<filename>"
Verify that you've used the correct filename of your gradio app, and that you're in the directory of the file.
ERR_PNPM_RECURSIVE_RUN_FIRST_FAIL @self/[email protected] build:local: vite build --mode production:local --emptyOutDir "--emptyOutDir"
Delete /node_modules
and pnpm-lock.yaml
:
rm -rf node_modules/
rm pnpm-lock.yaml
and run the install scripts:
bash scripts/install_gradio.sh
bash scripts/build_frontend.sh
FATAL ERROR: Ineffective mark-compacts near heap limit Allocation failed - JavaScript heap out of memory
when running scripts/build_frontend.sh
.
Run scripts/build_frontend.sh
with the environment variable NODE_OPTIONS=--max_old_space_size=2048
to increase the heap size.
In the case of:
- Unexpected exceptions being thrown, or
- The following warning:
IMPORTANT: You are using gradio version <earlier version>, however version <later version> is available, please upgrade.
ensure your PYTHONPATH
includes the directory where the Gradio repository is cloned, e.g.:
export PYTHONPATH="./"
This ensures that when gradio
is imported in a python program, it is this current version from this repository.
Could these guidelines be clearer? Feel free to open a PR to help us facilitate open-source contributions!