This repo includes the source code for the NVTabular C++ Triton Backend.
nvcr.io/nvstaging/merlin/merlin-inference
container includes all the required
packages and libraries to build the C++ backend from source. Please follow the
steps below;
$ conda install -c conda-forge rapidjson pybind11 cmake
$ git clone https://github.com/NVIDIA-Merlin/nvtabular_triton_backend.git
$ cd nvtabular_triton_backend/
$ mkdir build
$ cd build
$ cmake ..
$ make -j 4
$ mkdir /opt/tritonserver/backends/nvtabular
$ cp libtriton_nvtabular.so /opt/tritonserver/backends/nvtabular/
The model that will use NVTabular C++ backend should have the following in the config.pbtxt file
backend: "nvtabular"
Note that 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/backend: -DTRITON_BACKEND_REPO_TAG=[tag]
- triton-inference-server/core: -DTRITON_CORE_REPO_TAG=[tag]
- triton-inference-server/common: -DTRITON_COMMON_REPO_TAG=[tag]