NVIDIA Triton Inference Server 2.71.0

Use these rules for NVIDIA Triton Inference Server 2.71.0, corresponding to NGC release 26.07, including its serving frontends, repositories, backends, shared memory, and deployment.

Release and runtime compatibility

• Reference the selected NGC image by digest and verify that the packaged server reports the release this deployment was reviewed against. Build custom components from the release branch that matches the image, including the backend, common, and core dependencies; do not let build defaults select main.
• Take the NGC image tag and digest from the NGC catalog or the Triton release notes rather than memory, leaving a placeholder that fails the deployment when you can't check, and keep client packages such as tritonclient and grpcio in a lock file and the image digest in deployment configuration that an update bot or a scheduled review keeps current. NVIDIA's bulletins direct Triton fixes to newer monthly releases, so move to one when a bulletin lists yours as affected. The 2.71.0 in this prompt's title names the release these rules were checked against, not a version to install.
• Match the standard container's Ubuntu 24.04, Python 3.12, CUDA 13.3 family, and TensorRT 11.1 runtime. For GPU serving, require compute capability 7.5 or newer and verify the actual image's driver compatibility; CUDA 13.x minor compatibility starts at driver branch 580, but PTX and newer features can require newer drivers.
• Check specialized TensorRT-LLM, vLLM, Jetson, and native distributions against their own shipped compatibility requirements. Treat the manylinux_2_28