Thanks for your interest in contributing to Tripy!

1. Clone the repository:

   git clone https://github.com/NVIDIA/TensorRT-Incubator.git

2. Launch the development container.

   • In most cases, you can pull an existing one:

     1. Log in to the registry. Use your GitHub username and a personal access token as the password:

        docker login ghcr.io/nvidia/tensorrt-incubator

     2. Pull and launch the container. From the tripy root directory, run:

        docker run --pull always --gpus all -it --cap-add=SYS_PTRACE -p 8080:8080 -v $(pwd):/tripy/ --rm ghcr.io/nvidia/tensorrt-incubator/nvtripy-dev

   • If you made changes to the container (e.g. changing Dockerfile or pyproject.toml), build it locally and launch it. From the tripy root directory, run:

     docker build -t nvtripy-dev .
     docker run --gpus all -it --cap-add=SYS_PTRACE -p 8080:8080 -v $(pwd):/tripy/ --rm nvtripy-dev:latest

   • If you are using Visual Studio Code, you can alternatively use the included .devcontainer configuration.

3. [Optional] Run a sanity check in the container:

   python3 -c "import nvtripy as tp; print(tp.ones((2, 3)))"

   You should see output like:

   tensor(
       [[1. 1. 1.]
       [1. 1. 1.]],
       dtype=float32, loc=gpu:0, shape=(2, 3)
   )
   

From the tripy root directory, run:

python3 -m pip install pre-commit
pre-commit install

We've written developer guides to help you understand the codebase:

• Architecture guide

• Adding New Operations

See the tests README for details on adding tests for your changes.

If you add/modify any public-facing interfaces, update the documentation - see this README.

1. Set up commit signing.

2. Tripy follows fork based developement.

   • Create a fork on GitHub. Make git aware of it with:

     git remote add fork <FORK_URL>

   • Create a feature branch. For example:

     git checkout -b my-feature-branch

   • After committing changes, push to your fork:

     git push fork

   • Create a pull request on GitHub with a brief description explaining the change.

3. Contributions you make should satisfy the developer certificate of origin:

Developer Certificate of Origin Version 1.1

Copyright (C) 2004, 2006 The Linux Foundation and its contributors.

Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed.

Developer's Certificate of Origin 1.1

By making a contribution to this project, I certify that:

(a) The contribution was created in whole or in part by me and I have the right to submit it under the open source license indicated in the file; or

(b) The contribution is based upon previous work that, to the best of my knowledge, is covered under an appropriate open source license and I have the right under that license to submit that work with modifications, whether created in whole or in part by me, under the same open source license (unless I am permitted to submit under a different license), as indicated in the file; or

(c) The contribution was provided directly to me by some other person who certified (a), (b) or (c) and I have not modified it.

(d) I understand and agree that this project and the contribution are public and that a record of the contribution (including all personal information I submit with it, including my sign-off) is maintained indefinitely and may be redistributed consistent with this project or the open source license(s) involved.

Tripy depends on MLIR-TensorRT for compilation and execution. The Tripy container includes a build of MLIR-TensorRT, but in some cases, you may want to test Tripy with a local build:

1. Build MLIR-TensorRT as per the instructions in the README.

2. Launch the container with mlir-tensorrt repository mapped for accessing wheels files; from the tripy root directory, run:

   docker run --gpus all -it --cap-add=SYS_PTRACE -p 8080:8080 -v $(pwd):/tripy/ -v $(pwd)/../mlir-tensorrt:/mlir-tensorrt  --rm nvtripy-dev:latest

3. Install MLIR-TensorRT wheels MLIR-TensorRT must be built against a specific version of TensorRT, but will be able to work with any ABI-compatible version at runtime.

   The Tripy container includes TensorRT. Follow these steps to confirm the TensorRT version and ensure compatibility with your TensorRT wheels:

   echo "$LD_LIBRARY_PATH" | grep -oP 'TensorRT-\K\d+\.\d+\.\d+\.\d+'

   Ensure the installed MLIR-TensorRT wheels have:

   • The same major version as the TensorRT version in the container.
   • A minor version equal to or higher than the version in the container.

   For example, to install Python 3.10.12 wheels compatible with TensorRT 10.2+, run:

   python3 -m pip install --force-reinstall /mlir-tensorrt/build/wheels/python3.10.12/trt102/**/*.whl

4. Verify everything works:

   python3 -c "import nvtripy as tp; print(tp.ones((2, 3)))"