Skip to content

Latest commit

 

History

History
166 lines (113 loc) · 4.73 KB

DEVELOPING.md

File metadata and controls

166 lines (113 loc) · 4.73 KB
Raw

Developing SparseML

SparseML is developed and tested using Python 3.8-3.11. To develop SparseML, you will also need the development dependencies and to follow the styling guidelines.

Here are some details to get started.

Basic Commands

Development Installation

git clone https://github.com/neuralmagic/sparseml.git
cd sparseml
python3 -m pip install -e "./[dev]"

This will clone the SparseML repo, install it, and install the development dependencies.

To develop framework specific features, you will also need the relevant framework packages. Those can be installed by adding the framework name to the install extras. Frameworks include torch, keras, and tensorflow_v1. For example:

python3 -m pip install -e "./[dev,torch]"

Note: Running all pytorch tests using make test TARGETS=torch, also requires torchvision and onnxruntime install all these dependencies using python3 -m pip install -e "./[dev, torch, torchvision, onnxruntime]"

Code Styling and Formatting checks

make style
make quality

This will run automatic code styling using black and isort and test that the repository's code matches its standards.

EXAMPLE: test changes locally

make test TARGETS=<CSV of frameworks to run>

This will run the targeted SparseML unit tests for the frameworks specified. The targets should be specified, because not all framework dependencies can be installed to run all tests.

To run just PyTorch tests, run

make test TARGETS=pytorch

File any error found before changes as an Issue and fix any errors found after making changes before submitting a Pull Request.

GitHub Workflow

  1. Fork the neuralmagic/sparseml repository into your GitHub account: https://github.com/neuralmagic/sparseml/fork.

  2. Clone your fork of the GitHub repository, replacing <username> with your GitHub username.

    Use ssh (recommended):

    git clone [email protected]:<username>/sparseml.git

    Or https:

    git clone https://github.com/<username>/sparseml.git

  3. Add a remote to keep up with upstream changes.

    git remote add upstream https://github.com/neuralmagic/sparseml.git

    If you already have a copy, fetch upstream changes.

    git fetch upstream

  4. Create a feature branch to work in.

    git checkout -b feature-xxx remotes/upstream/main

  5. Work in your feature branch.

    git commit -a

  6. Periodically rebase your changes

    git pull --rebase

  7. When done, combine ("squash") related commits into a single one

    git rebase -i upstream/main

    This will open your editor and allow you to re-order commits and merge them:

    • Re-order the lines to change commit order (to the extent possible without creating conflicts)
    • Prefix commits using s (squash) or f (fixup) to merge extraneous commits.

  8. Submit a pull-request

    git push origin feature-xxx

    Go to your fork main page

    https://github.com/<username>/sparseml

    If you recently pushed your changes GitHub will automatically pop up a Compare & pull request button for any branches you recently pushed to. If you click that button it will automatically offer you to submit your pull-request to the neuralmagic/sparseml repository.

    • Give your pull-request a meaningful title. You'll know your title is properly formatted once the Semantic Pull Request GitHub check transitions from a status of "pending" to "passed".
    • In the description, explain your changes and the problem they are solving.

  9. Addressing code review comments

    Repeat steps 5. through 7. to address any code review comments and rebase your changes if necessary.

    Push your updated changes to update the pull request

    git push origin [--force] feature-xxx

    --force may be necessary to overwrite your existing pull request in case your commit history was changed when performing the rebase.

    Note: Be careful when using --force since you may lose data if you are not careful.

    git push origin --force feature-xxx