This repository holds the source code and headers for the library that implements the core functionality of Triton. The core library can be built as described below and used directly via its C API. To be useful the core library must be paired with one or more backends. You can learn more about backends in the backend repo.
Typically you do not build or use the core library on its own, but as part of the tritonserver executable. The tritonserver executable is built in the server repo as described in the server build documentation.
Ask questions or report problems in the main Triton issues page.
Before building the Triton core library, your build system must install the required dependencies described in the build documentation. For example, if you are building the core library with GPU support (-DTRITON_ENABLE_GPU=ON), then you must install the CUDA, cuDNN, and TensorRT dependencies required for the version of Triton you are building.
To build, first clone the release branch matching the Triton release you are interest in (rxx.yy), or the main branch to build the top-of-tree. The Triton core library is built with CMake.
$ mkdir build
$ cd build
$ cmake -DCMAKE_INSTALL_PREFIX:PATH=`pwd`/install -DTRITON_CORE_HEADERS_ONLY=OFF ..
$ make install
When the build completes, the install directory will contain the Triton core shared library (install/lib/libtritonserver.so on Linux, install/bin/tritonserver.dll on Windows), and the core library headers files in install/include/triton/core.
The following required Triton repositories will be pulled and used in the build. By default the "main" branch/tag will be used for each repo but the listed CMake argument can be used to override.
- triton-inference-server/third_party: -DTRITON_THIRD_PARTY_REPO_TAG=[tag]
- triton-inference-server/common: -DTRITON_COMMON_REPO_TAG=[tag]
You will need to override if you are building from a release branch. For example, if you are building the r22.03 version of Triton, you would clone the r22.03 branch of the core repo and use the following cmake command.
$ cmake -DTRITON_THIRD_PARTY_REPO_TAG=r22.03 -DTRITON_COMMON_REPO_TAG=r22.03 -DTRITON_CORE_HEADERS_ONLY=OFF ..
The CMakeLists.txt file contains the options available when build the core library. For example, to build the core library with the default settings plus S3 cloud storage and ensembling support use the following command.
$ cmake -DTRITON_CORE_HEADERS_ONLY=OFF -DTRITON_ENABLE_S3=ON -DTRITON_ENABLE_ENSEMBLE=ON ..