[TRTLLM-14027][infra] Remove --trt_root and stop installing the TensorRT SDK into images

.claude/skills/exec-local-compile/SKILL.md

@@ -45,7 +45,7 @@ git checkout main && git pull
 Run the build command (**incremental by default** — omit `-c`/`--clean` unless explicitly requested or the incremental build fails):
 
 ```bash
-./scripts/build_wheel.py --trt_root /usr/local/tensorrt --use_ccache -a "<arch>" -f --nvtx
+./scripts/build_wheel.py --use_ccache -a "<arch>" -f --nvtx
 ```
 
 Replace `<arch>` with the target GPU architecture (see Architecture Reference below). If not specified by the user, auto-detect from `nvidia-smi`.
@@ -66,7 +66,6 @@ python3 -c "import tensorrt_llm; print(tensorrt_llm.__version__)"
 
 | Flag | Description |
 |------|-------------|
-| `--trt_root /usr/local/tensorrt` | TensorRT installation path (standard in NVIDIA containers) |
 | `-a "<arch>"` | Target GPU architecture(s) |
 | `--nvtx` | Enable NVTX markers for profiling |
 | `--use_ccache` | Use ccache for faster recompilation |

.claude/skills/exec-slurm-compile/SKILL.md

@@ -204,7 +204,6 @@ A successful build ends with a message like `Successfully built tensorrt_llm` or
 
 | Flag | Description |
 |------|-------------|
-| `--trt_root /usr/local/tensorrt` | TensorRT installation path (standard in NVIDIA containers) |
 | `-a "100-real"` | Target architecture — `100` for Blackwell, `90` for Hopper, etc. |
 | `--nvtx` | Enable NVTX markers for profiling |
 | `--no-venv` | Skip virtual environment creation |
@@ -227,7 +226,6 @@ Common architecture values:
 | `sbatch: error: invalid partition` | Verify partition name with `sinfo -s` |
 | `sbatch: error: invalid account` | Check available accounts with `sacctmgr show assoc user=$USER` |
 | Container image not found | Verify the `.sqsh` path exists and is readable |
-| Build fails with missing TensorRT | Ensure `--trt_root` points to the correct path inside the container |
 | Build OOM (out of memory) | Reduce parallelism with `-j <N>` flag to `build_wheel.py` |
 | `srun: error: Unable to create step` | The node may lack enroot/pyxis — check with cluster admin |
 | Job stuck in `PD` state | Check `squeue -j <id> -o %R` for the reason (e.g., resource limits, priority) |

.claude/skills/exec-slurm-compile/scripts/compile.sh

@@ -19,7 +19,7 @@
 # Usage: compile.sh <repo_dir> [build_wheel_args...]
 #
 # Default build_wheel.py flags:
-#   --trt_root /usr/local/tensorrt -a "100-real" --nvtx --no-venv
+#   -a "100-real" --nvtx --no-venv
 # Any extra arguments after repo_dir are forwarded to build_wheel.py,
 # overriding the defaults above.
 
@@ -36,7 +36,6 @@ if [[ $# -gt 0 ]]; then
 else
     echo "[compile.sh] Running default build command"
     python3 ./scripts/build_wheel.py \
-        --trt_root /usr/local/tensorrt \
         -a "100-real" \
         --nvtx
 fi

docker/Dockerfile.multi

@@ -31,7 +31,6 @@ FROM base AS devel
 # NB: PyTorch requires this to be < 1.0
 ENV PYTORCH_ALLOC_CONF="garbage_collection_threshold:0.99999"
 
-ARG TRT_VER
 ARG CUDA_VER
 ARG CUDNN_VER
 ARG NCCL_VER
@@ -45,11 +44,11 @@ RUN --mount=type=bind,source=docker/common,target=/opt/docker/common \
     echo "Using Python version: ${PYTHON_VERSION}" && \
     GITHUB_MIRROR=${GITHUB_MIRROR} \
     PYTHON_VERSION=${PYTHON_VERSION} \
-    TRT_VER=${TRT_VER} CUDA_VER=${CUDA_VER} CUDNN_VER=${CUDNN_VER} \
+    CUDA_VER=${CUDA_VER} CUDNN_VER=${CUDNN_VER} \
     NCCL_VER=${NCCL_VER} CUBLAS_VER=${CUBLAS_VER} \
     TORCH_INSTALL_TYPE=${TORCH_INSTALL_TYPE} \
     bash /opt/docker/common/install.sh --base --cmake --ccache --cuda_toolkit \
-         --tensorrt --polygraphy --mpi4py --pytorch
+         --cuda_libs --polygraphy --mpi4py --pytorch
 
 # Install constraints after install.sh so cleanup() doesn't delete the file mid-RUN
 COPY constraints.txt /tmp/constraints.txt

docker/Makefile

@@ -47,7 +47,6 @@ CUDA_VERSION       ?=
 CUDNN_VERSION      ?=
 NCCL_VERSION       ?=
 CUBLAS_VERSION     ?=
-TRT_VERSION        ?=
 GIT_COMMIT         ?= $(shell git rev-parse HEAD)
 TRT_LLM_VERSION    ?= $(shell grep '^__version__' ../tensorrt_llm/version.py | grep -o '=.*' | tr -d '= "')
 GITHUB_MIRROR      ?=
@@ -99,7 +98,6 @@ base_pull:
 		$(if $(CUDNN_VERSION), --build-arg CUDNN_VER="$(CUDNN_VERSION)") \
 		$(if $(NCCL_VERSION), --build-arg NCCL_VER="$(NCCL_VERSION)") \
 		$(if $(CUBLAS_VERSION), --build-arg CUBLAS_VER="$(CUBLAS_VERSION)") \
-		$(if $(TRT_VERSION), --build-arg TRT_VER="$(TRT_VERSION)") \
 		$(if $(TRT_LLM_VERSION), --build-arg TRT_LLM_VER="$(TRT_LLM_VERSION)") \
 		$(if $(DEVEL_IMAGE), --build-arg DEVEL_IMAGE="$(DEVEL_IMAGE)") \
 		$(if $(GIT_COMMIT), --build-arg GIT_COMMIT="$(GIT_COMMIT)") \

docker/common/install.sh

@@ -11,7 +11,7 @@ base=0
 cmake=0
 ccache=0
 cuda_toolkit=0
-tensorrt=0
+cuda_libs=0
 polygraphy=0
 mpi4py=0
 pytorch=0
@@ -34,8 +34,8 @@ while [[ $# -gt 0 ]]; do
             cuda_toolkit=1
             shift 1
             ;;
-        --tensorrt)
-            tensorrt=1
+        --cuda_libs)
+            cuda_libs=1
             shift 1
             ;;
         --polygraphy)
@@ -55,7 +55,7 @@ while [[ $# -gt 0 ]]; do
             cmake=1
             ccache=1
             cuda_toolkit=1
-            tensorrt=1
+            cuda_libs=1
             polygraphy=1
             mpi4py=1
             pytorch=1
@@ -92,10 +92,9 @@ if [ $cuda_toolkit -eq 1 ]; then
     bash $SCRIPT_DIR/install_cuda_toolkit.sh
 fi
 
-if [ $tensorrt -eq 1 ]; then
-    echo "Installing TensorRT..."
-    bash $SCRIPT_DIR/install_tensorrt.sh \
-        --TRT_VER=${TRT_VER} \
+if [ $cuda_libs -eq 1 ]; then
+    echo "Installing CUDA libraries (cuDNN/NCCL/cuBLAS)..."
+    bash $SCRIPT_DIR/install_cuda_libs.sh \
         --CUDA_VER=${CUDA_VER} \
         --CUDNN_VER=${CUDNN_VER} \
         --NCCL_VER=${NCCL_VER} \

docker/common/install_tensorrt.sh → docker/common/install_cuda_libs.sh

@@ -2,7 +2,6 @@
 
 set -ex
 
-TRT_VER="10.16.1.11"
 # Align with the pre-installed cuDNN / cuBLAS / NCCL versions from
 # https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-26-05.html#rel-26-05
 CUDA_VER="13.2" # 13.2.1
@@ -19,7 +18,6 @@ CUDA_DRIVER_VERSION="595.58.03-1.el8"
 
 for i in "$@"; do
     case $i in
-        --TRT_VER=?*) TRT_VER="${i#*=}";;
         --CUDA_VER=?*) CUDA_VER="${i#*=}";;
         --CUDNN_VER=?*) CUDNN_VER="${i#*=}";;
         --NCCL_VER=?*) NCCL_VER="${i#*=}";;
@@ -166,47 +164,14 @@ install_rockylinux_requirements() {
     ldconfig
 }
 
-install_tensorrt() {
-    PY_VERSION=$(python3 -c 'import sys; print(".".join(map(str, sys.version_info[0:2])))')
-    PARSED_PY_VERSION=$(echo "${PY_VERSION//./}")
-
-    TRT_CUDA_VERSION=${CUDA_VER}
-    TRT_VER_SHORT=$(echo $TRT_VER | cut -d. -f1-3)
-
-    if [ -z "$RELEASE_URL_TRT" ];then
-        ARCH=${TRT_TARGETARCH}
-        if [ -z "$ARCH" ];then ARCH=$(uname -m);fi
-        if [ "$ARCH" = "arm64" ];then ARCH="aarch64";fi
-        if [ "$ARCH" = "amd64" ];then ARCH="x86_64";fi
-        RELEASE_URL_TRT="https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/${TRT_VER_SHORT}/tars/TensorRT-${TRT_VER}.Linux.${ARCH}-gnu.cuda-${TRT_CUDA_VERSION}.tar.gz"
-    fi
-
-    wget --retry-connrefused --timeout=180 --tries=10 --continue ${RELEASE_URL_TRT} -O /tmp/TensorRT.tar
-    tar -xf /tmp/TensorRT.tar -C /usr/local/
-    mv /usr/local/TensorRT-${TRT_VER} /usr/local/tensorrt
-    pip3 install --no-cache-dir /usr/local/tensorrt/python/tensorrt-*-cp${PARSED_PY_VERSION}-*.whl
-    rm -rf /tmp/TensorRT.tar
-    echo 'export LD_LIBRARY_PATH=/usr/local/tensorrt/lib:$LD_LIBRARY_PATH' >> "${ENV}"
-
-    rm -f /usr/local/tensorrt/lib/libnvinfer_vc_plugin_static.a \
-          /usr/local/tensorrt/lib/libnvinfer_plugin_static.a \
-          /usr/local/tensorrt/lib/libnvinfer_static.a \
-          /usr/local/tensorrt/lib/libnvinfer_dispatch_static.a \
-          /usr/local/tensorrt/lib/libnvinfer_lean_static.a \
-          /usr/local/tensorrt/lib/libnvonnxparser_static.a \
-          /usr/local/tensorrt/lib/libnvinfer_builder_resource_win.so.*
-}
-
 # Install base packages depending on the base OS
 ID=$(grep -oP '(?<=^ID=).+' /etc/os-release | tr -d '"')
 case "$ID" in
   ubuntu)
     install_ubuntu_requirements
-    install_tensorrt
     ;;
   rocky)
     install_rockylinux_requirements
-    install_tensorrt
     ;;
   *)
     echo "Unable to determine OS..."

jenkins/L0_Test.groovy

@@ -4428,9 +4428,12 @@ def runPackageSanityCheck(pipeline, wheel_path, reinstall_dependencies=false, cp
     trtllm_utils.llmExecStepWithRetry(pipeline, script: "wget -nv ${pkgUrl}")
     sh "tar -zvxf ${linuxPkgName}"
 
-    trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core/gpt && python3 ../../../generate_checkpoint_config.py --architecture GPTForCausalLM --dtype float16'")
-    trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core//gpt && trtllm-build --model_config config.json --log_level verbose'")
-    trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core/gpt && python3 ../../../run.py --max_output_len 4 --end_id -1'")
+    // TODO: The steps below drove the removed TensorRT engine flow (trtllm-build / examples/run.py).
+    // When re-enabling this sanity check, use PyTorch backend test samples instead
+    // (e.g. examples/llm-api/quickstart_example.py).
+    // trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core/gpt && python3 ../../../generate_checkpoint_config.py --architecture GPTForCausalLM --dtype float16'")
+    // trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core//gpt && trtllm-build --model_config config.json --log_level verbose'")
+    // trtllm_utils.llmExecStepWithRetry(pipeline, script: "bash -c 'cd tensorrt_llm/examples/models/core/gpt && python3 ../../../run.py --max_output_len 4 --end_id -1'")
 }
 
 def checkStageNameSet(stageNames, jobKeys, paramName) {

jenkins/current_image_tags.properties

@@ -13,8 +13,8 @@
 #     images are adopted from PostMerge pipelines, the abbreviated commit hash is used instead.
 IMAGE_NAME=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm
 
-LLM_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-26.05-py3-x86_64-ubuntu24.04-trt10.16.1.11-skip-tritondevel-202607151440-16194
-LLM_SBSA_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-26.05-py3-sbsa-ubuntu24.04-trt10.16.1.11-skip-tritondevel-202607151440-16194
-LLM_ROCKYLINUX8_PY310_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-rocky8-x86_64-rocky8-py310-trt10.16.1.11-skip-tritondevel-202607151440-16194
-LLM_ROCKYLINUX8_PY312_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-rocky8-x86_64-rocky8-py312-trt10.16.1.11-skip-tritondevel-202607151440-16194
-LLM_SBSA_WHEEL_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-ubuntu24.04-sbsa-ubuntu24.04-py312-trt10.16.1.11-skip-tritondevel-202607151440-16194
+LLM_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-26.05-py3-x86_64-ubuntu24.04-skip-tritondevel-202607211045-16608
+LLM_SBSA_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:pytorch-26.05-py3-sbsa-ubuntu24.04-skip-tritondevel-202607211045-16608
+LLM_ROCKYLINUX8_PY310_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-rocky8-x86_64-rocky8-py310-skip-tritondevel-202607211045-16608
+LLM_ROCKYLINUX8_PY312_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-rocky8-x86_64-rocky8-py312-skip-tritondevel-202607211045-16608
+LLM_SBSA_WHEEL_DOCKER_IMAGE=urm.nvidia.com/sw-tensorrt-docker/tensorrt-llm:cuda-13.2.1-devel-ubuntu24.04-sbsa-ubuntu24.04-py312-skip-tritondevel-202607211045-16608

scripts/build_wheel.py

@@ -489,7 +489,6 @@ def main(*,
          job_count: int = None,
          extra_cmake_vars: Sequence[str] = tuple(),
          extra_make_targets: str = "",
-         trt_root: str = None,
          nccl_root: str = None,
          nixl_root: str = None,
          mooncake_root: str = None,
@@ -596,9 +595,6 @@ def main(*,
         # Don't include duplicate conditions
         cmake_def_args.extend(set(extra_cmake_vars))
 
-    if trt_root is not None:
-        cmake_def_args.append(f"-DTensorRT_ROOT={trt_root}")
-
     if nccl_root is not None:
         cmake_def_args.append(f"-DNCCL_ROOT={nccl_root}")
 
@@ -1002,11 +998,6 @@ def copy_resolving_symlink(src_path, dst_path):
     scripts_dir = pkg_dir / "scripts"
     if scripts_dir.exists():
         clear_folder(scripts_dir)
-    scripts_dir.mkdir(parents=True, exist_ok=True)
-
-    if not on_windows:
-        install_file(project_dir / "docker/common/install_tensorrt.sh",
-                     scripts_dir / "install_tensorrt.sh")
 
     if not cpp_only:
 
@@ -1235,12 +1226,6 @@ def add_arguments(parser: ArgumentParser):
         help="Additional make targets to build. Example: \"target_1 target_2\"",
         nargs="+",
         default=[])
-    parser.add_argument(
-        "--trt_root",
-        default="/usr/local/tensorrt",
-        help="[DEPRECATED] No effect: TensorRT is no longer required to build. "
-        "Accepted for backward compatibility and will be removed in a future release."
-    )
     parser.add_argument("--nccl_root",
                         help="Directory containing NCCL headers and libraries")
     parser.add_argument("--nixl_root",

setup.py

@@ -166,7 +166,6 @@ def has_ext_modules(self):
         'deep_gemm/include/**/*',
         'deep_gemm/*.py',
         'deep_gemm_cpp_tllm.*.so',
-        'scripts/install_tensorrt.sh',
         'flash_mla/LICENSE',
         'flash_mla/*.py',
         'flash_mla_cpp_tllm.*.so',

tests/integration/defs/cpp/conftest.py

@@ -167,7 +167,6 @@ def build_google_tests(request, build_type):
         use_ccache=True,
         clean=True,
         generator="Ninja",
-        trt_root="/usr/local/tensorrt",
         nixl_root="/opt/nvidia/nvda_nixl",
         skip_building_wheel=True,
         extra_make_targets=["google-tests"],