From 69880f62b0c40914891a55f893e7ef36ff76db18 Mon Sep 17 00:00:00 2001 From: Mohammed Sheikh Date: Thu, 6 Aug 2026 12:20:33 +0530 Subject: [PATCH] Upgraded Triton v2.70.0 --- t/triton-inference-server/fil_backend.patch | 106 ++++++- .../onnxruntime_backend.patch | 40 +-- .../pytorch_backend.patch | 30 +- t/triton-inference-server/rhelppc.patch | 290 ++++++++++++------ .../triton-inference-server-master-rhel9.6.sh | 10 +- 5 files changed, 338 insertions(+), 138 deletions(-) diff --git a/t/triton-inference-server/fil_backend.patch b/t/triton-inference-server/fil_backend.patch index 969e1bb1d9..f7b4002946 100644 --- a/t/triton-inference-server/fil_backend.patch +++ b/t/triton-inference-server/fil_backend.patch @@ -1,16 +1,116 @@ diff --git a/CMakeLists.txt b/CMakeLists.txt -index 9aff54c..6ea6a77 100644 +index f14add1..0a7eb88 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt -@@ -78,6 +78,10 @@ else() +@@ -78,7 +78,10 @@ else() # - Target names ------------------------------------------------------------- set(BACKEND_NAME "fil") set(BACKEND_TARGET "triton_${BACKEND_NAME}") +- + install( + FILES ${CMAKE_CURRENT_BINARY_DIR}/libtriton_fil.so + DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/fil + ) + ############################################################################## + # - Prepare rapids-cmake ----------------------------------------------------- +@@ -274,7 +277,19 @@ else() + TARGETS ${BACKEND_TARGET} + LIBRARY DESTINATION ${BACKEND_FOLDER}/${BACKEND_NAME} + ) ++ # nvforest++ is a CPM sub-project target; install(TARGETS) cannot reach it ++ # from the top-level project. Install the built shared library directly. ++ install( ++ FILES $ ++ DESTINATION ${BACKEND_FOLDER}/${BACKEND_NAME} ++ ) ++ # rapids_logger is a header-only / shared lib that nvforest++ links publicly. ++ # It must sit next to libtriton_fil.so so the $ORIGIN RPATH resolves it. ++ install( ++ FILES $ ++ DESTINATION ${BACKEND_FOLDER}/${BACKEND_NAME} ++ ) ############################################################################## - + # - build test executable ---------------------------------------------------- + +diff --git a/ops/Dockerfile b/ops/Dockerfile +index e4b4115..18defa2 100644 +--- a/ops/Dockerfile ++++ b/ops/Dockerfile +@@ -46,14 +46,21 @@ RUN gpuci_mamba_retry create -n triton_test \ + && rm /environment.yml + + FROM base-test-install as wheel-install-0 +-RUN apt-get update \ +- && apt-get install --no-install-recommends -y \ +- build-essential \ +- ca-certificates \ +- git \ +- && apt-get clean \ +- && rm -rf /var/lib/apt/lists/* \ +- && conda run --no-capture-output -n triton_test pip install tritonclient[all] ++#RUN apt-get update \ ++# && apt-get install --no-install-recommends -y \ ++# build-essential \ ++# ca-certificates \ ++# git \ ++# && apt-get clean \ ++# && rm -rf /var/lib/apt/lists/* \ ++# && conda run --no-capture-output -n triton_test pip install tritonclient[all] ++RUN yum install -y \ ++ gcc \ ++ gcc-c++ \ ++ make \ ++ patch \ ++ ca-certificates \ ++ git + + FROM ${SDK_IMAGE} as sdk-image + +@@ -80,22 +87,29 @@ ENV PATH="/root/miniconda3/bin:${PATH}" + + # In CI, CPU base image may not have curl, but it also does not need to update + # the cuda keys +-RUN if command -v curl; \ +- then [ $(uname -m) = 'x86_64' ] \ +- && curl -L -o /tmp/cuda-keyring.deb \ +- https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.0-1_all.deb \ +- || curl -L -o /tmp/cuda-keyring.deb \ +- https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/sbsa/cuda-keyring_1.0-1_all.deb; \ +- dpkg -i /tmp/cuda-keyring.deb \ +- && rm /tmp/cuda-keyring.deb; fi +- +-RUN apt-get update \ +- && apt-get install --no-install-recommends -y \ +- build-essential \ +- ca-certificates \ +- git \ +- && apt-get clean \ +- && rm -rf /var/lib/apt/lists/* ++#RUN if command -v curl; \ ++# then [ $(uname -m) = 'x86_64' ] \ ++# && curl -L -o /tmp/cuda-keyring.deb \ ++# https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.0-1_all.deb \ ++# || curl -L -o /tmp/cuda-keyring.deb \ ++# https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/sbsa/cuda-keyring_1.0-1_all.deb; \ ++# dpkg -i /tmp/cuda-keyring.deb \ ++# && rm /tmp/cuda-keyring.deb; fi ++ ++#RUN apt-get update \ ++# && apt-get install --no-install-recommends -y \ ++# build-essential \ ++# ca-certificates \ ++# git \ ++# && apt-get clean \ ++# && rm -rf /var/lib/apt/lists/* ++RUN yum install -y \ ++ gcc \ ++ gcc-c++ \ ++ make \ ++ patch \ ++ ca-certificates \ ++ git + + # Stage immediately before building; useful for build iteration + FROM base as build-prep diff --git a/t/triton-inference-server/onnxruntime_backend.patch b/t/triton-inference-server/onnxruntime_backend.patch index 2c984ec5e5..9b6420a263 100644 --- a/t/triton-inference-server/onnxruntime_backend.patch +++ b/t/triton-inference-server/onnxruntime_backend.patch @@ -23,20 +23,14 @@ index bf374cd..c53c6b9 100644 COMMAND docker create --name onnxruntime_backend_ort ${TRITON_ONNXRUNTIME_DOCKER_IMAGE} COMMAND rm -fr onnxruntime diff --git a/tools/gen_ort_dockerfile.py b/tools/gen_ort_dockerfile.py -index c462558..c3d58a9 100755 +index cdb4253..dde34db 100755 --- a/tools/gen_ort_dockerfile.py +++ b/tools/gen_ort_dockerfile.py -@@ -134,27 +134,31 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 +@@ -170,17 +170,27 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 + # if the changes become more substantial. + if target_platform() == "rhel": df += """ - # The manylinux container defaults to Python 3.7, but some feature installation - # requires a higher version. --ARG PYVER=3.12 --ENV PYTHONPATH=/opt/python/v --RUN ln -sf /opt/python/cp${PYVER/./}* ${PYTHONPATH} - --ENV PYBIN=${PYTHONPATH}/bin --ENV PYTHON_BIN_PATH=${PYBIN}/python${PYVER} \ -- PATH=${PYBIN}:${PATH} +- +RUN yum install -y wget && \ + dnf install -y https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-gpg-keys-9.0-35.el9.noarch.rpm \ + https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-stream-repos-9.0-35.el9.noarch.rpm \ @@ -44,29 +38,23 @@ index c462558..c3d58a9 100755 + dnf config-manager --set-enabled crb &&\ + dnf install gperf libarchive-devel numactl-devel readline-devel libusb patch -y && \ + dnf remove -y centos-gpg-keys-9.0-35.el9.noarch centos-stream-repos-9.0-35.el9.noarch - RUN dnf install -y \\ ca-certificates \\ - curl \\ git \\ - gnupg \\ openssl-devel \\ -- python3-pip \\ ++ python3.12 \\ + python3.12-devel \\ + python3.12-pip \\ ++ python3.12-numpy \\ wget \\ - zip + zip \\ -+ python3.12 \\ -+ python3.12-pip \\ -+ python3.12-devel \\ -+ python3.12-numpy \\ -+ cmake \\ -+ g++ - --RUN pipx install cmake==4.0.3 --force ++ cmake \\ ++ g++ ++ +RUN ln -sf $(which python3.12) /usr/bin/python3 && ln -sf $(which pip3.12) /usr/bin/pip3 && ln -sf $(which pip3.12) /usr/bin/pip -+RUN pip3 install patchelf==0.17.2 cmake==4.0.3 packaging wheel --RUN pip3 install patchelf==0.17.2 numpy>=2.0.0 - """ - else: - if os.getenv("CCACHE_REMOTE_ONLY") and os.getenv("CCACHE_REMOTE_STORAGE"): + RUN pip3 install \\ + cmake==4.0.3 \\ diff --git a/t/triton-inference-server/pytorch_backend.patch b/t/triton-inference-server/pytorch_backend.patch index 916df700a6..cd8c3d51a1 100644 --- a/t/triton-inference-server/pytorch_backend.patch +++ b/t/triton-inference-server/pytorch_backend.patch @@ -1,8 +1,8 @@ diff --git a/CMakeLists.txt b/CMakeLists.txt -index 13c32d9..b4d1c74 100644 +index 49536a6..3ed3360 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt -@@ -90,8 +90,9 @@ find_package(Python3 REQUIRED COMPONENTS Development.Module) +@@ -91,8 +91,9 @@ find_package(Python3 REQUIRED COMPONENTS Interpreter Development.Module) set(RHEL_BUILD OFF) set(LIB_DIR "lib") @@ -10,11 +10,11 @@ index 13c32d9..b4d1c74 100644 -set(PY_INSTALL_PATH "/usr/local/lib/python3.12/dist-packages") +set(LIBTORCH_LIBS_PATH "/usr/local/lib64") +set(PY_INSTALL_PATH "/usr/local/lib64/python3.12/site-packages") -+message(WARNING "USING UBUNTU INSTEAD OF RHEL") ++message(WARNING "USING RHEL INSTEAD OF UBUNTU") if(LINUX) file(STRINGS "/etc/os-release" DISTRO_ID_LIKE REGEX "ID_LIKE") if(${DISTRO_ID_LIKE} MATCHES "rhel|centos") -@@ -175,26 +176,16 @@ configure_file(src/libtriton_pytorch.ldscript libtriton_pytorch.ldscript COPYONL +@@ -218,27 +219,16 @@ configure_file(src/libtriton_pytorch.ldscript libtriton_pytorch.ldscript COPYONL set(PT_LIBS "libc10.so" @@ -38,10 +38,11 @@ index 13c32d9..b4d1c74 100644 - "libtorch_nvshmem.so" - ) -endif() # TRITON_PYTORCH_NVSHMEM - +- if (${TRITON_PYTORCH_ENABLE_TORCHVISION}) set(PT_LIBS -@@ -226,19 +217,6 @@ if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64") + ${PT_LIBS} +@@ -271,19 +261,6 @@ if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64") "libnvpl_lapack_lp64_seq.so.0" ) else() @@ -60,8 +61,8 @@ index 13c32d9..b4d1c74 100644 - ) endif() set(TORCHVISION_LIBS - $,libjpeg.so.62,libjpeg.so> -@@ -269,41 +247,15 @@ if (${TRITON_PYTORCH_DOCKER_BUILD}) + libjpeg.so.62 +@@ -314,41 +291,15 @@ if (${TRITON_PYTORCH_DOCKER_BUILD}) COMMAND docker create --name pytorch_backend_ptlib ${TRITON_PYTORCH_DOCKER_IMAGE} COMMAND /bin/sh -c "for i in ${LIBTORCH_LIBS_STR} ; do echo copying $i && docker cp -L pytorch_backend_ptlib:${LIBTORCH_LIBS_PATH}/$i $i ; done" COMMAND docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch/lib/libc10.so libc10.so @@ -78,8 +79,7 @@ index 13c32d9..b4d1c74 100644 COMMAND docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch/lib/libtorch_python.so libtorch_python.so - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_NVSHMEM} = 'ON' ]; then docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch/lib/libtorch_nvshmem.so libtorch_nvshmem.so; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then docker cp -a -L pytorch_backend_ptlib:/usr/local/${LIB_DIR}/libtorchvision.so.1 libtorchvision.so.1; fi;" -+ COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -a -L pytorch_backend_ptlib:/usr/local/lib64/libtorchvision.so libtorchvision.so; else docker cp -a -L pytorch_backend_ptlib:/usr/local/lib64/libtorchvision.so.1 libtorchvision.so.1; fi; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then docker cp pytorch_backend_ptlib:/opt/pytorch/vision/torchvision/csrc include/torchvision/torchvision; fi" +- COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then docker cp pytorch_backend_ptlib:/opt/pytorch/vision/torchvision/csrc include/torchvision/torchvision; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHTRT} = 'ON' ]; then docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch_tensorrt/lib/libtorchtrt.so libtorchtrt.so; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHTRT} = 'ON' ]; then docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch_tensorrt/lib/libtorchtrt_runtime.so libtorchtrt_runtime.so; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHTRT} = 'ON' ]; then docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch_tensorrt/bin/torchtrtc torchtrtc; fi" @@ -87,8 +87,8 @@ index 13c32d9..b4d1c74 100644 - COMMAND docker cp pytorch_backend_ptlib:${PY_INSTALL_PATH}/torch/include include/torch - COMMAND docker cp pytorch_backend_ptlib:/opt/pytorch/pytorch/torch/csrc/jit/codegen include/torch/torch/csrc/jit/. - -- COMMAND /bin/sh -c "if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -L pytorch_backend_ptlib:/usr/lib64/libjpeg.so.62 libjpeg.so.62; else docker cp -L pytorch_backend_ptlib:/usr/local/lib/libjpeg.so.62 libjpeg.so.62 && docker cp pytorch_backend_ptlib:/usr/lib/${LIBS_ARCH}-linux-gnu/libjpeg.so.8.2.2 libjpeg.so; fi;" -- COMMAND /bin/sh -c "if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -L pytorch_backend_ptlib:/usr/lib64/libpng16.so.16 libpng16.so.16; else docker cp -L pytorch_backend_ptlib:/usr/lib/${LIBS_ARCH}-linux-gnu/libpng16.so libpng16.so; fi;" +- COMMAND /bin/sh -c "if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -L pytorch_backend_ptlib:/usr/lib64/libjpeg.so.62 libjpeg.so.62; else docker cp -L pytorch_backend_ptlib:/usr/local/lib/libjpeg.so.62 libjpeg.so.62 ; fi;" +- COMMAND /bin/sh -c "if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -L pytorch_backend_ptlib:/usr/lib64/libpng16.so.16 libpng16.so.16; else docker cp -L pytorch_backend_ptlib:/usr/lib/${LIBS_ARCH}-linux-gnu/libpng16.so.16 libpng16.so.16; fi;" - COMMAND /bin/sh -c "if [ -f libmkl_def.so.1 ]; then patchelf --add-needed libmkl_gnu_thread.so.1 libmkl_def.so.1; fi" - COMMAND /bin/sh -c "if [ -f libmkl_def.so.1 ]; then patchelf --add-needed libmkl_core.so.1 libmkl_def.so.1; fi" - COMMAND /bin/sh -c "if [ -f libmkl_avx2.so.1 ]; then patchelf --add-needed libmkl_gnu_thread.so.1 libmkl_avx2.so.1; fi" @@ -101,14 +101,16 @@ index 13c32d9..b4d1c74 100644 - COMMAND /bin/sh -c "if [ -f libmkl_intel_thread.so.1 ]; then patchelf --add-needed libmkl_intel_lp64.so.1 libmkl_intel_thread.so.1; fi" - COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then ln -s libtorchvision.so.1 libtorchvision.so; fi;" - COMMAND docker rm pytorch_backend_ptlib ++ COMMAND /bin/sh -c "if [ ${TRITON_PYTORCH_ENABLE_TORCHVISION} = 'ON' ]; then if [ ${RHEL_BUILD} = 'ON' ]; then docker cp -a -L pytorch_backend_ptlib:/usr/local/lib64/torchvision.so libtorchvision.so; else edocker cp -a -L pytorch_backend_ptlib:/usr/local/lib64/libtorchvision.so.1 libtorchvision.so.1; fi; fi" ++ COMMENT "Extracting pytorch and torchvision libraries and includes from ${TRITON_PYTORCH_DOCKER_IMAGE}" VERBATIM ) diff --git a/src/libtorch_utils.h b/src/libtorch_utils.h -index 6ec325b..0ac84bc 100644 +index 4ced3dc..481dac1 100644 --- a/src/libtorch_utils.h +++ b/src/libtorch_utils.h -@@ -34,7 +34,6 @@ +@@ -38,7 +38,6 @@ #pragma GCC diagnostic ignored "-Wsign-compare" #pragma warning(push, 0) #include diff --git a/t/triton-inference-server/rhelppc.patch b/t/triton-inference-server/rhelppc.patch index 1c84190a7f..5c8b963615 100644 --- a/t/triton-inference-server/rhelppc.patch +++ b/t/triton-inference-server/rhelppc.patch @@ -1,8 +1,8 @@ diff --git a/build.py b/build.py -index f32c0159..77718cc7 100755 +index 326b443d..11a833f5 100755 --- a/build.py +++ b/build.py -@@ -335,6 +335,13 @@ class BuildScript: +@@ -282,6 +282,13 @@ class BuildScript: f" git clone --recursive --single-branch --depth=1 -b {tag} {org}/{repo}.git {subdir}; git --git-dir {subdir}/.git log --oneline -1", check_exitcode=True, ) @@ -13,10 +13,10 @@ index f32c0159..77718cc7 100755 + self.cmd(f"git apply /workspace/pytorch_backend.patch", check_exitcode=True) + if subdir == "fil": + self.cmd(f"git apply /workspace/fil_backend.patch", check_exitcode=True) - self.cmd("}" if target_platform() == "windows" else "fi") + self.cmd("fi") -@@ -457,7 +464,8 @@ def core_cmake_args(components, backends, cmake_dir, install_dir): +@@ -404,7 +411,8 @@ def core_cmake_args(components, backends, cmake_dir, install_dir): "TRITON_THIRD_PARTY_REPO_TAG", "STRING", components["thirdparty"] ), ] @@ -26,7 +26,7 @@ index f32c0159..77718cc7 100755 cargs.append(cmake_core_enable("TRITON_ENABLE_LOGGING", FLAGS.enable_logging)) cargs.append(cmake_core_enable("TRITON_ENABLE_STATS", FLAGS.enable_stats)) cargs.append(cmake_core_enable("TRITON_ENABLE_METRICS", FLAGS.enable_metrics)) -@@ -649,6 +657,9 @@ def pytorch_cmake_args(images): +@@ -589,6 +597,9 @@ def pytorch_cmake_args(images): cargs = [ cmake_backend_arg("pytorch", "TRITON_PYTORCH_DOCKER_IMAGE", None, image), ] @@ -36,30 +36,28 @@ index f32c0159..77718cc7 100755 # TODO: TPRD-372 TorchTRT extension is not currently supported by our manylinux build # TODO: TPRD-373 NVTX extension is not currently supported by our manylinux build -@@ -935,11 +946,20 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 CMAKE_POLICY_VERSION_MINIMUM=3.5 +@@ -853,10 +864,18 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 CMAKE_POLICY_VERSION_MINIMUM=3.5 """ df += """ # Install docker docker buildx -RUN yum install -y ca-certificates curl gnupg yum-utils \\ - && yum-config-manager --add-repo https://download.docker.com/linux/rhel/docker-ce.repo \\ -+RUN yum update -y && yum install -y ca-certificates gnupg yum-utils \\ ++RUN yum update -y && yum install -y ca-certificates gnupg yum-utils \\ + && yum-config-manager --add-repo https://download.docker.com/linux/centos/docker-ce.repo \\ && yum install -y docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin # && yum install -y docker.io docker-buildx-plugin - +RUN yum update -y && yum install -y wget && \ -+ dnf install -y https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-gpg-keys-9.0-35.el9.noarch.rpm \ ++ dnf install -y https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-gpg-keys-9.0-35.el9.noarch.rpm \ + https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-stream-repos-9.0-35.el9.noarch.rpm \ + https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm && \ + dnf config-manager --set-enabled crb &&\ + dnf install gperf libarchive-devel numactl-devel readline-devel -y && \ + dnf install --nodocs -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm &&\ + dnf remove -y centos-gpg-keys-9.0-35.el9.noarch centos-stream-repos-9.0-35.el9.noarch -+ + # libcurl4-openSSL-dev is needed for GCS # python3-dev is needed by Torchvision - # python3-pip and libarchive-dev is needed by python backend -@@ -948,12 +968,15 @@ RUN yum install -y ca-certificates curl gnupg yum-utils \\ +@@ -866,12 +885,15 @@ RUN yum install -y ca-certificates curl gnupg yum-utils \\ RUN yum install -y \\ autoconf \\ automake \\ @@ -75,7 +73,7 @@ index f32c0159..77718cc7 100755 libb64-devel \\ libcurl-devel \\ libtool \\ -@@ -962,16 +985,33 @@ RUN yum install -y \\ +@@ -880,16 +902,103 @@ RUN yum install -y \\ numactl-devel \\ openssl-devel \\ pkg-config \\ @@ -94,26 +92,101 @@ index f32c0159..77718cc7 100755 - zlib-devel + zlib-devel \\ + libjpeg-devel \\ -+ sqlite-devel\\ -+ libffi-devel\\ ++ sqlite-devel \\ ++ libffi-devel \\ + openblas-devel \\ + g++ \\ + ccache \\ -+ gfortran ++ gfortran \\ ++ patch \\ ++ ninja-build +RUN ln -sf $(which python3.12) /usr/bin/python3 && ln -sf $(which pip3.12) /usr/bin/pip3 && ln -sf $(which pip3.12) /usr/bin/pip +RUN ln -sf $(which python3.12) /usr/bin/python3 && ln -sf $(which pip3.12) /usr/bin/pip3 && ln -sf $(which pip3.12) /usr/bin/pip ++RUN wget https://wheels.developerfirst.ibm.com/ppc64le/linux/+f/564/0ed0c96830382/torch-2.9.0+ppc64le3-cp312-cp312-manylinux_2_34_ppc64le.whl#sha256=5640ed0c96830382037e88a81718dbc0a7a9445be371404b1724d018d794d87a && pip3 install torch-2.9.0+ppc64le3-cp312-cp312-manylinux_2_34_ppc64le.whl ++RUN wget https://wheels.developerfirst.ibm.com/ppc64le/linux/+f/6bc/8735aba99d30f/torchvision-0.24.1+ppc64le2-cp312-cp312-manylinux_2_34_ppc64le.whl#sha256=6bc8735aba99d30f1c01d8c860cb3cfda877ce3a7b4c5cd6c84de722429462bf && pip3 install torchvision-0.24.1+ppc64le2-cp312-cp312-manylinux_2_34_ppc64le.whl ++# Copy libtorchvision.so and system libs into the torch lib directory so the ++# linker and CMake install rules find them at build time. ++# libjpeg.so.62: shipped by the OS (libjpeg-turbo) but expected by the pytorch ++# backend CMakeLists.txt install rule from within torch/lib. ++RUN cp /usr/local/lib64/python3.12/site-packages/torchvision/libtorchvision.so \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so && \\ ++ ln -sf /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so.1 && \\ ++ cp /usr/lib64/libjpeg.so.62 \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libjpeg.so.62 ++ ++# ------------------------------------------------------------------ ++# Build prerequisites for protobuf ++# ------------------------------------------------------------------ ++RUN echo "-------------------- abseil-cpp installing -------------------------------" && \ ++ export C_COMPILER=$(which gcc) && \ ++ export CXX_COMPILER=$(which g++) && \ ++ cd / && \ ++ git clone https://github.com/abseil/abseil-cpp -b 20240116.2 && \ ++ echo "------------ abseil-cpp cloned --------------" ++ ++# ------------------------------------------------------------------ ++# Build and install protobuf 4.25.8 from source ++# ------------------------------------------------------------------ ++RUN echo "---------------- protobuf installing ----------------" && \ ++ cd / && \ ++ git clone https://github.com/protocolbuffers/protobuf && \ ++ cd protobuf && \ ++ git checkout v4.25.8 && \ ++ git submodule update --init --recursive && \ ++ rm -rf ./third_party/googletest || true && \ ++ rm -rf ./third_party/abseil-cpp || true ++ RUN cp -r /abseil-cpp /protobuf/third_party/abseil-cpp ++ + -+RUN wget https://raw.githubusercontent.com/ppc64le/build-scripts/refs/heads/master/t/torchvision/torchvision_ubi_9.3.sh && bash torchvision_ubi_9.3.sh ++RUN cd /protobuf && \ ++ mkdir -p local/libprotobuf && \ ++ mkdir -p build + -+RUN cp /vision/build/libtorchvision.so /usr/local/lib64/python3.12/site-packages/torch/lib/ ++RUN cd /protobuf/build && \ ++ cmake -G Ninja \ ++ -DCMAKE_BUILD_TYPE=Release \ ++ -DCMAKE_CXX_STANDARD=17 \ ++ -DCMAKE_INSTALL_PREFIX=/protobuf/local/libprotobuf \ ++ -Dprotobuf_BUILD_TESTS=OFF \ ++ -Dprotobuf_BUILD_LIBUPB=OFF \ ++ -Dprotobuf_BUILD_SHARED_LIBS=ON \ ++ -Dprotobuf_ABSL_PROVIDER=module \ ++ -Dprotobuf_JSONCPP_PROVIDER=package \ ++ -Dprotobuf_USE_EXTERNAL_GTEST=OFF \ ++ .. && \ ++ echo "building libprotobuf...." && \ ++ cmake --build . --verbose && \ ++ echo "Installing libprotobuf...." && \ ++ cmake --install . + -+RUN ln -s /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so.1 -+RUN cp /usr/lib64/libjpeg.so.62 /usr/local/lib64/python3.12/site-packages/torch/lib/ -+RUN wget https://cmake.org/files/v3.31/cmake-3.31.8.tar.gz && tar -zxvf cmake-3.31.8.tar.gz && cd cmake-3.31.8 && ./bootstrap && make && make install && cd .. ++ENV LIBPROTO_INSTALL=/protobuf/local/libprotobuf ++ENV PROTOC=/protobuf/build/protoc ++ ++ENV LD_LIBRARY_PATH=/protobuf/local/libprotobuf/lib64:${LD_LIBRARY_PATH} ++ENV LIBRARY_PATH=/protobuf/local/libprotobuf/lib64:${LIBRARY_PATH} ++ENV PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=cpp ++ENV PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION_VERSION=2 ++ ++RUN cd /protobuf && \ ++ wget https://raw.githubusercontent.com/ppc64le/build-scripts/refs/heads/master/p/protobuf/set_cpp_to_17_v4.25.3.patch && \ ++ git apply set_cpp_to_17_v4.25.3.patch ++RUN pip3 install --upgrade pip "setuptools==78.1.1" wheel build ++RUN cd /protobuf/python && \ ++ python3 -m pip install . --no-build-isolation ++ ++RUN echo "------------ libprotobuf, protobuf installed --------------" ++ ++RUN wget https://github.com/Kitware/CMake/releases/download/v4.0.4/cmake-4.0.4.tar.gz && tar -zxvf cmake-4.0.4.tar.gz && cd cmake-4.0.4 && ./bootstrap && make && make install && cd .. """ - if os.getenv("CCACHE_REMOTE_ONLY") and os.getenv("CCACHE_REMOTE_STORAGE"): + if argmap["NVIDIA_BUILD_ID"] is not None: df += """ -@@ -1003,7 +1043,7 @@ RUN pip3 install --upgrade pip \\ +@@ -925,11 +1034,11 @@ RUN pip3 install --upgrade pip \\ + auditwheel \\ + build \\ + wheel \\ +- setuptools \\ ++ "setuptools==78.1.1" \\ docker \\ virtualenv \\ patchelf==0.17.2 \\ @@ -122,89 +195,96 @@ index f32c0159..77718cc7 100755 """ df += f""" # Install boost version >= 1.78 for boost::span -@@ -1115,8 +1155,7 @@ RUN pip3 install --upgrade \\ - build \\ +@@ -1041,7 +1150,6 @@ RUN pip3 install --upgrade \\ docker \\ virtualenv \\ -- patchelf==0.17.2 \\ -- cmake==4.0.3 -+ patchelf==0.17.2 + patchelf==0.17.2 \\ +- cmake==4.0.3 \\ + pybind11[global] """ - df += f""" -@@ -1218,16 +1257,9 @@ ARG BASE_IMAGE={} +@@ -1144,16 +1252,11 @@ ARG TRITON_CONTAINER_VERSION={} # PyTorch backends need extra CUDA and other # dependencies during runtime that are missing in the CPU-only base container. # These dependencies must be copied from the Triton Min image. - if not FLAGS.enable_gpu and ("pytorch" in backends): - df += """ ++ ############################################################################ ## Triton Min image ############################################################################ -FROM {} AS min_container -- + -""".format( - argmap["GPU_BASE_IMAGE"] - ) df += """ ############################################################################ -@@ -1246,15 +1278,38 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 +@@ -1167,10 +1270,28 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1 + df += dockerfile_prepare_container_linux( + argmap, backends, FLAGS.enable_gpu, target_machine() + ) +- ++ variant = "cpu" if target_machine() == "ppc64le" else FLAGS.triton_wheels_dependencies_group + df += f""" WORKDIR /opt - COPY --chown=1000:1000 build/install tritonserver - -+COPY build/torch-2.6.0-cp312-cp312-linux_ppc64le.whl /opt -+COPY build/torchvision-0.22.0-cp312-cp312-linux_ppc64le.whl /opt + COPY build/install tritonserver +RUN mkdir protobuf +COPY build/protobuf /opt/protobuf -+ +RUN cp /opt/protobuf/build/* /usr/lib64 -r && cp /opt/protobuf/local/libprotobuf/lib64/* /usr/lib64 -r && rm -rf /opt/protobuf -+RUN pip3 install /opt/torch-2.6.0-cp312-cp312-linux_ppc64le.whl /opt/torchvision-0.22.0-cp312-cp312-linux_ppc64le.whl -+RUN rm -rf /opt/torch-2.6.0-cp312-cp312-linux_ppc64le.whl /opt/torchvision-0.22.0-cp312-cp312-linux_ppc64le.wh -+COPY build/libtorchvision.so /usr/local/lib64/python3.12/site-packages/torch/share/cmake/Torch -+COPY build/libtorchvision.so /usr/local/lib64/python3.12/site-packages/torch/lib -+RUN ln -s /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so.1 -+ ++# Install torch and torchvision wheels copied out of the builder container. ++# The final UBI9 image has no torch/torchvision pre-installed, so we install ++# them here and then stage libtorchvision.so where tritonserver expects it. ++COPY build/torch-2.9.0+ppc64le3-cp312-cp312-manylinux_2_34_ppc64le.whl /tmp/ ++COPY build/torchvision-0.24.1+ppc64le2-cp312-cp312-manylinux_2_34_ppc64le.whl /tmp/ ++RUN pip3 install /tmp/torch-2.9.0+ppc64le3-cp312-cp312-manylinux_2_34_ppc64le.whl && \\ ++ pip3 install /tmp/torchvision-0.24.1+ppc64le2-cp312-cp312-manylinux_2_34_ppc64le.whl && \\ ++ rm -f /tmp/torch-*.whl /tmp/torchvision-*.whl ++# Stage libtorchvision.so and libjpeg.so.62 into the tritonserver pytorch backend ++RUN cp /usr/local/lib64/python3.12/site-packages/torchvision/libtorchvision.so \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so && \\ ++ ln -sf /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libtorchvision.so.1 && \\ ++ cp /usr/lib64/libjpeg.so.62 \\ ++ /usr/local/lib64/python3.12/site-packages/torch/lib/libjpeg.so.62 + WORKDIR /opt/tritonserver - COPY --chown=1000:1000 NVIDIA_Deep_Learning_Container_License.pdf . --RUN find /opt/tritonserver/python -maxdepth 1 -type f -name \\ -- "tritonserver-*.whl" | xargs -I {{}} pip install --upgrade {{}}[{FLAGS.triton_wheels_dependencies_group}] && \\ -- find /opt/tritonserver/python -maxdepth 1 -type f -name \\ -- "tritonfrontend-*.whl" | xargs -I {{}} pip install --upgrade {{}}[{FLAGS.triton_wheels_dependencies_group}] -+RUN if [ "$(uname -m)" = "ppc64le" ]; then \ -+ VARIANT="cpu"; \ -+ else \ -+ VARIANT="all"; \ -+ fi &&\ -+ find /opt/tritonserver/python -maxdepth 1 -type f -name \ -+ "tritonserver-*.whl" | xargs -I {{}} pip install --upgrade {{}}[$VARIANT] && \ -+ find /opt/tritonserver/python -maxdepth 1 -type f -name \ -+ "tritonfrontend-*.whl" | xargs -I {{}} pip install --upgrade {{}}[$VARIANT]; + COPY NVIDIA_Deep_Learning_Container_License.pdf . +@@ -1187,7 +1308,7 @@ RUN set -e; \\ + echo "ERROR: ${{pkg}}-*.whl missing from /opt/tritonserver/python -- build did not stage the wheel into the image" >&2; \\ + exit 1; \\ + fi; \\ +- printf '%s\\n' "$wheels" | xargs -I {{}} pip install --upgrade "{{}}[{FLAGS.triton_wheels_dependencies_group}]"; \\ ++ printf '%s\\n' "$wheels" | xargs -I {{}} pip install --upgrade "{{}}[{variant}]"; \\ + done RUN pip3 install -r python/openai/requirements.txt - -+ENV LD_LIBRARY_PATH=/opt/tritonserver/lib64:/usr/local/lib64/python3.12/site-packages/torch/lib:$LD_LIBRARY_PATH -+ -+LABEL maintainer="Sunidhi Gaonkar" \ -+ version="v2.65.0" \ -+ description="Triton Inference Server is an open source inference serving software that streamlines AI inferencing.This CPU-only image enables teams to deploy any AI model from multiple deep learning and machine learning frameworks,including PyTorch,ONNX,Python,and FIL." -+ - """ - if not FLAGS.no_core_build: - # Add feature labels for SageMaker endpoint -@@ -1268,7 +1323,6 @@ COPY --chown=1000:1000 docker/sagemaker/serve /usr/bin/. +@@ -1205,7 +1326,7 @@ COPY docker/sagemaker/serve /usr/bin/. # stage of the PyTorch backend if not FLAGS.enable_gpu and ("pytorch" in backends): df += """ -RUN patchelf --add-needed /usr/local/cuda/lib64/stubs/libcublasLt.so.13 backends/pytorch/libtorch_cuda.so ++ """ if "tensorrtllm" in backends: df += """ -@@ -1352,21 +1406,45 @@ RUN userdel tensorrt-server > /dev/null 2>&1 || true \\ +@@ -1234,6 +1355,11 @@ ENV PATH /opt/tritonserver/bin:${PATH} + # Remove once https://github.com/openucx/ucx/pull/9148 is available + # in the min container. + ENV UCX_MEM_EVENTS no ++""" ++ ++ if target_platform() == "rhel": ++ df += """ ++ENV LD_LIBRARY_PATH /opt/tritonserver/lib64:${LD_LIBRARY_PATH} + """ + + # Necessary for libtorch.so to find correct HPCX libraries +@@ -1279,21 +1405,44 @@ RUN userdel tensorrt-server > /dev/null 2>&1 || true \\ if target_platform() == "rhel": df += """ -+ +RUN yum update -y && yum install -y wget && \ + dnf install -y https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-gpg-keys-9.0-35.el9.noarch.rpm \ + https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-stream-repos-9.0-35.el9.noarch.rpm \ @@ -213,7 +293,6 @@ index f32c0159..77718cc7 100755 + dnf install gperf numactl-devel readline-devel -y && \ + dnf install --nodocs -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm &&\ + dnf remove -y centos-gpg-keys-9.0-35.el9.noarch centos-stream-repos-9.0-35.el9.noarch -+ + # Common dependencies. RUN yum install -y \\ @@ -230,13 +309,10 @@ index f32c0159..77718cc7 100755 + civetweb-devel \\ + libpng-devel \\ wget \\ -- python3-pip \\ -- numactl-devel -- --RUN pip3 install patchelf==0.17.2 + cmake \\ + python3.12 \\ -+ python3.12-pip \\ + python3.12-pip \\ +- numactl-devel + python3.12-devel \\ + numactl-devel \\ + libjpeg-devel \\ @@ -246,11 +322,13 @@ index f32c0159..77718cc7 100755 + ccache \\ + gfortran +RUN ln -sf $(which python3.12) /usr/bin/python3 && ln -sf $(which pip3.12) /usr/bin/pip3 && ln -sf $(which pip3.12) /usr/bin/pip -+RUN pip3 install patchelf==0.17.2 pillow==12.2.0 ++RUN pip3 install patchelf==0.17.2 pillow==12.3.0 + +-RUN pip3 install patchelf==0.17.2 """ else: -@@ -1435,7 +1513,6 @@ RUN ln -sf ${_CUDA_COMPAT_PATH}/lib.real ${_CUDA_COMPAT_PATH}/lib \\ +@@ -1363,7 +1512,6 @@ RUN ln -sf ${_CUDA_COMPAT_PATH}/lib.real ${_CUDA_COMPAT_PATH}/lib \\ df += """ # python3, python3-pip and some pip installs required for the python backend RUN yum install -y \\ @@ -258,19 +336,28 @@ index f32c0159..77718cc7 100755 openssl-devel \\ readline-devel """ -@@ -1522,11 +1599,6 @@ ENV PYTHONPATH=/opt/tritonserver/backends/dali/wheel/dali:$PYTHONPATH +@@ -1373,7 +1521,7 @@ RUN yum install -y \\ + RUN pip3 install --upgrade pip \\ + && pip3 install --upgrade \\ + wheel \\ +- setuptools \\ ++ "setuptools==78.1.1" \\ + \"numpy<2\" \\ + virtualenv + """ +@@ -1412,11 +1560,7 @@ ENV PYTHONPATH=/opt/tritonserver/backends/dali/wheel/dali:$PYTHONPATH if target_platform() == "rhel": repo_arch = "sbsa" if target_machine == "aarch64" else "x86_64" df += """ -RUN dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/{repo_arch}/cuda-rhel8.repo \\ - && dnf clean expire-cache \\ - && dnf install --assumeyes libnvshmem3-cuda-13 -- + -RUN dirname $(find /usr -name "libcudart*.so" -o -name "libnvinf*.so" -o -name "libnvshm*" -type f) | sort -u > /etc/ld.so.conf.d/triton-cuda-libs.conf && ldconfig """.format( repo_arch=repo_arch ) -@@ -1568,38 +1640,9 @@ def add_cpu_libs_to_linux_dockerfile(backends, target_machine): +@@ -1458,38 +1602,12 @@ def add_cpu_libs_to_linux_dockerfile(backends, target_machine): # we must copy these from the Triton min container ourselves. cuda_arch = "sbsa" if target_machine == "aarch64" else "x86_64" df += """ @@ -299,25 +386,28 @@ index f32c0159..77718cc7 100755 -COPY --from=min_container /opt/hpcx/ucx/lib/libuct.so.0 /opt/hpcx/ucx/lib/libuct.so.0 - -COPY --from=min_container /usr/lib/{libs_arch}-linux-gnu/libcudnn.so.9 /usr/lib/{libs_arch}-linux-gnu/libcudnn.so.9 -- ++ + # patchelf is needed to add deps of libcublasLt.so.12 to libtorch_cuda.so -RUN apt-get update \\ - && apt-get install -y --no-install-recommends openmpi-bin RUN pip3 install patchelf==0.17.2 -ENV LD_LIBRARY_PATH /usr/local/cuda/targets/{cuda_arch}-linux/lib:/usr/local/cuda/lib64/stubs:${{LD_LIBRARY_PATH}} ++ """.format( cuda_arch=cuda_arch, libs_arch=libs_arch ) -@@ -1611,7 +1654,6 @@ ENV LD_LIBRARY_PATH /usr/local/cuda/targets/{cuda_arch}-linux/lib:/usr/local/cud +@@ -1501,7 +1619,7 @@ ENV LD_LIBRARY_PATH /usr/local/cuda/targets/{cuda_arch}-linux/lib:/usr/local/cud # Since this dependency is not present in the ubuntu base image, # we must copy it from the Triton min container ourselves. df += """ -COPY --from=min_container /usr/lib/{libs_arch}-linux-gnu/libnccl.so.2 /usr/lib/{libs_arch}-linux-gnu/libnccl.so.2 ++ """.format( libs_arch=libs_arch ) -@@ -1625,20 +1667,6 @@ def change_default_python_version_rhel(version): +@@ -1515,20 +1633,7 @@ def change_default_python_version_rhel(version): # match the version of python inside the RHEL base container. This means that python packages # installed within the container will not be picked up by the python backend stub process pybind # bindings. It must instead must be installed via pyenv. @@ -335,18 +425,19 @@ index f32c0159..77718cc7 100755 -RUN ln -sf ${{PYENV_ROOT}}/versions/${{PYVER}}* ${{PYTHONPATH}} -ENV PYBIN=${{PYTHONPATH}}/bin -ENV PYTHON_BIN_PATH=${{PYBIN}}/python${{PYVER}} PATH=${{PYBIN}}:${{PATH}} ++ """ return df -@@ -1896,6 +1924,43 @@ def create_docker_build_script(script_name, container_install_dir, container_ci_ +@@ -1727,6 +1832,43 @@ def create_docker_build_script(script_name, container_install_dir, container_ci_ + ], check_exitcode=True, ) - + docker_script.cmd( + [ + "docker", + "cp", -+ "tritonserver_builder:/torch-2.6.0-cp312-cp312-linux_ppc64le.whl", ++ "tritonserver_builder:/torch-2.9.0+ppc64le3-cp312-cp312-manylinux_2_34_ppc64le.whl", + FLAGS.build_dir, + ], + check_exitcode=True, @@ -355,7 +446,7 @@ index f32c0159..77718cc7 100755 + [ + "docker", + "cp", -+ "tritonserver_builder:/torchvision-0.22.0-cp312-cp312-linux_ppc64le.whl", ++ "tritonserver_builder:/torchvision-0.24.1+ppc64le2-cp312-cp312-manylinux_2_34_ppc64le.whl", + FLAGS.build_dir, + ], + check_exitcode=True, @@ -364,7 +455,7 @@ index f32c0159..77718cc7 100755 + [ + "docker", + "cp", -+ "tritonserver_builder:/vision/build/libtorchvision.so", ++ "tritonserver_builder:/usr/local/lib64/python3.12/site-packages/torchvision/libtorchvision.so", + FLAGS.build_dir, + ], + check_exitcode=True, @@ -379,9 +470,28 @@ index f32c0159..77718cc7 100755 + check_exitcode=True, + ) + + # # Final image... tritonserver - # +@@ -1938,6 +2080,18 @@ def backend_build( + if be == "tensorrtllm": + tensorrtllm_be_dir = os.path.join(build_dir, be) + tensorrtllm_postbuild(cmake_script, repo_install_dir, tensorrtllm_be_dir) ++ # The FIL backend's CMakeLists.txt installs companion shared libraries ++ # (libnvforest++.so, librapids_logger.so) directly into ++ # /opt/tritonserver/backends/fil/ via a hardcoded BACKEND_FOLDER, rather ++ # than into CMAKE_INSTALL_PREFIX/backends/fil. Sync them into ++ # repo_install_dir so the cpdir below stages them into the final image. ++ if be == "fil" and target_platform() == "rhel": ++ fil_system_dir = "/opt/tritonserver/backends/fil" ++ fil_stage_dir = os.path.join(repo_install_dir, "backends", be) ++ cmake_script.mkdir(fil_stage_dir) ++ cmake_script.cmd( ++ f"cp -n {fil_system_dir}/*.so {fil_stage_dir}/ 2>/dev/null || true" ++ ) + + cmake_script.mkdir(os.path.join(install_dir, "backends")) + cmake_script.rmdir(os.path.join(install_dir, "backends", be)) diff --git a/python/openai/requirements.txt b/python/openai/requirements.txt index 4b4f631f..9d614e77 100644 --- a/python/openai/requirements.txt diff --git a/t/triton-inference-server/triton-inference-server-master-rhel9.6.sh b/t/triton-inference-server/triton-inference-server-master-rhel9.6.sh index 89c3ecfe4d..fcdc6b231a 100644 --- a/t/triton-inference-server/triton-inference-server-master-rhel9.6.sh +++ b/t/triton-inference-server/triton-inference-server-master-rhel9.6.sh @@ -2,13 +2,13 @@ # ---------------------------------------------------------------------------- # # Package : triton-inference-server -# Version : v2.65.0 +# Version : v2.70.0 # Source repo : https://github.com/triton-inference-server/server -# Tested on : RHEL 9.6 +# Tested on : RHEL 9.8 # Language : Python , Shell # Ci-Check : False # Script License: Apache License, Version 2 or later -# Maintainer : Sunidhi Gaonkar +# Maintainer : Mohammed Sheikh # # Disclaimer: This script has been tested in root mode on given # ========== platform using the mentioned version of the package. @@ -23,7 +23,7 @@ wdir=`pwd` PACKAGE_NAME=server PACKAGE_URL=https://github.com/triton-inference-server/server -PACKAGE_VERSION=${1:-v2.65.0} +PACKAGE_VERSION=${1:-v2.70.0} yum install git python3.12-devel python3.12-pip cmake -y @@ -40,7 +40,7 @@ cp $wdir/pytorch_backend.patch . cp $wdir/fil_backend.patch . git apply $wdir/rhelppc.patch -if ! ./build.py --enable-logging --enable-metrics --enable-stats --endpoint http --backend onnxruntime --backend python --backend pytorch --backend fil --override-backend-cmake-arg=fil:TRITON_FIL_DOCKER_BUILD=OFF --image base,registry.access.redhat.com/ubi9/ubi:9.6 ; then +if ! ./build.py --enable-logging --enable-metrics --enable-stats --endpoint http --backend onnxruntime --backend python --backend pytorch --backend fil --override-backend-cmake-arg=fil:TRITON_FIL_DOCKER_BUILD=OFF --image base,registry.access.redhat.com/ubi9/ubi:9.8 ; then echo "------------------$PACKAGE_NAME:Build_fails---------------------" exit 2 else