From b46baea99bd661e5dd54013e39243c3554c8c8ce Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Wed, 24 Sep 2025 14:42:37 -0700 Subject: [PATCH 1/9] debug windows --- .github/workflows/build_windows.yml | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/.github/workflows/build_windows.yml b/.github/workflows/build_windows.yml index c2d4b0b20b..a780524b73 100644 --- a/.github/workflows/build_windows.yml +++ b/.github/workflows/build_windows.yml @@ -346,6 +346,11 @@ jobs: source "${BUILD_ENV_FILE}" WHEEL_NAME=$(ls "${{ inputs.repository }}/dist/") echo "$WHEEL_NAME" + set -x + nvidia-smi + nvcc --version + ${CONDA_RUN} python -m pip list + ${CONDA_RUN} pip install "${{ inputs.repository }}/dist/$WHEEL_NAME" if [[ $USE_TRT_RTX == true ]]; then # TODO: lan to remove this once we have a better way to do a smoke test From 593a22767bec3369d1a582863ec69f968a7c4efb Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Wed, 24 Sep 2025 15:16:31 -0700 Subject: [PATCH 2/9] debug windows --- .github/workflows/build-test-windows.yml | 2 ++ .github/workflows/build_windows.yml | 15 ++++++++------- 2 files changed, 10 insertions(+), 7 deletions(-) diff --git a/.github/workflows/build-test-windows.yml b/.github/workflows/build-test-windows.yml index 39c3d20571..b97601a0a2 100644 --- a/.github/workflows/build-test-windows.yml +++ b/.github/workflows/build-test-windows.yml @@ -138,6 +138,8 @@ jobs: pre-script: packaging/driver_upgrade.bat script: | set -euo pipefail + nvidia-smi + nvcc --version export USE_HOST_DEPS=1 export CI_BUILD=1 pushd . diff --git a/.github/workflows/build_windows.yml b/.github/workflows/build_windows.yml index a780524b73..07f4de9cb6 100644 --- a/.github/workflows/build_windows.yml +++ b/.github/workflows/build_windows.yml @@ -356,13 +356,14 @@ jobs: # TODO: lan to remove this once we have a better way to do a smoke test echo "Smoke test for TensorRT-RTX is not skipped for now" else - if [[ ! -f "${{ inputs.repository }}"/${SMOKE_TEST_SCRIPT} ]]; then - echo "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT} not found" - ${CONDA_RUN} "${{ inputs.repository }}/${ENV_SCRIPT}" python -c "import ${PACKAGE_NAME}; print('package version is ', ${PACKAGE_NAME}.__version__)" - else - echo "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT} found" - ${CONDA_RUN} "${{ inputs.repository }}/${ENV_SCRIPT}" python "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT}" - fi + echo "Skip smoke test for windows." + # if [[ ! -f "${{ inputs.repository }}"/${SMOKE_TEST_SCRIPT} ]]; then + # echo "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT} not found" + # ${CONDA_RUN} "${{ inputs.repository }}/${ENV_SCRIPT}" python -c "import ${PACKAGE_NAME}; print('package version is ', ${PACKAGE_NAME}.__version__)" + # else + # echo "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT} found" + # ${CONDA_RUN} "${{ inputs.repository }}/${ENV_SCRIPT}" python "${{ inputs.repository }}/${SMOKE_TEST_SCRIPT}" + # fi fi - name: Smoke Test ARM64 if: inputs.architecture == 'arm64' From b047e26d52b2e17c5016332a220d112151701490 Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Wed, 24 Sep 2025 13:48:35 -0700 Subject: [PATCH 3/9] fix test case error From c32f06b8909fe001b9250bbffb7ecd548c43d42d Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Thu, 25 Sep 2025 14:20:20 -0700 Subject: [PATCH 4/9] update tensorrt tar link according to cu13* or cu12* --- MODULE.bazel | 12 ++++++------ dev_dep_versions.yml | 2 +- packaging/pre_build_script.sh | 7 +++++++ packaging/pre_build_script_windows.sh | 6 ++++++ toolchains/ci_workspaces/MODULE.bazel.tmpl | 12 ++++++------ 5 files changed, 26 insertions(+), 13 deletions(-) diff --git a/MODULE.bazel b/MODULE.bazel index 12682f2186..f59a71267b 100644 --- a/MODULE.bazel +++ b/MODULE.bazel @@ -101,9 +101,9 @@ http_archive( http_archive( name = "tensorrt", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-12.9.tar.gz", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.x86_64-gnu.cuda-13.0.tar.gz", ], ) @@ -119,9 +119,9 @@ http_archive( http_archive( name = "tensorrt_sbsa", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.aarch64-gnu.cuda-13.0.tar.gz", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.aarch64-gnu.cuda-13.0.tar.gz", ], ) @@ -137,9 +137,9 @@ http_archive( http_archive( name = "tensorrt_win", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/zip/TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/zip/TensorRT-10.13.3.9.Windows.win10.cuda-13.0.zip", ], ) diff --git a/dev_dep_versions.yml b/dev_dep_versions.yml index 113fe23de6..1159951385 100644 --- a/dev_dep_versions.yml +++ b/dev_dep_versions.yml @@ -1,3 +1,3 @@ __cuda_version__: "12.8" -__tensorrt_version__: "10.12.0" +__tensorrt_version__: "10.13.3" __tensorrt_rtx_version__: "1.0.0" diff --git a/packaging/pre_build_script.sh b/packaging/pre_build_script.sh index 32b91ff3fe..cb2b9a1dd9 100755 --- a/packaging/pre_build_script.sh +++ b/packaging/pre_build_script.sh @@ -70,6 +70,13 @@ if [[ ${TENSORRT_VERSION} != "" ]]; then pyproject.toml fi +# CU_UPPERBOUND eg:13.0 or 12.9 +if [[ ${CU_VERSION:2:2} == "13" ]]; then + CU_UPPERBOUND="13.0" +else: + CU_UPPERBOUND="12.9" +fi + cat toolchains/ci_workspaces/MODULE.bazel.tmpl | envsubst > MODULE.bazel if [[ ${TENSORRT_VERSION} != "" ]]; then diff --git a/packaging/pre_build_script_windows.sh b/packaging/pre_build_script_windows.sh index 4be0018f0d..696e90c511 100644 --- a/packaging/pre_build_script_windows.sh +++ b/packaging/pre_build_script_windows.sh @@ -27,6 +27,12 @@ pip install --force-reinstall --pre ${TORCH} --index-url ${INDEX_URL} export CUDA_HOME="$(echo ${CUDA_PATH} | sed -e 's#\\#\/#g')" export TORCH_INSTALL_PATH="$(python -c "import torch, os; print(os.path.dirname(torch.__file__))" | sed -e 's#\\#\/#g')" +# CU_UPPERBOUND eg:13.0 or 12.9 +if [[ ${CU_VERSION:2:2} == "13" ]]; then + CU_UPPERBOUND="13.0" +else: + CU_UPPERBOUND="12.9" +fi cat toolchains/ci_workspaces/MODULE.bazel.tmpl | envsubst > MODULE.bazel if [[ ${TENSORRT_VERSION} != "" ]]; then diff --git a/toolchains/ci_workspaces/MODULE.bazel.tmpl b/toolchains/ci_workspaces/MODULE.bazel.tmpl index 491d5f4ac3..0a68df8370 100644 --- a/toolchains/ci_workspaces/MODULE.bazel.tmpl +++ b/toolchains/ci_workspaces/MODULE.bazel.tmpl @@ -75,9 +75,9 @@ http_archive = use_repo_rule("@bazel_tools//tools/build_defs/repo:http.bzl", "ht http_archive( name = "tensorrt", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-12.9.tar.gz", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.x86_64-gnu.cuda-${CU_UPPERBOUND}.tar.gz", ], ) @@ -93,9 +93,9 @@ http_archive( http_archive( name = "tensorrt_sbsa", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.aarch64-gnu.cuda-13.0.tar.gz", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.aarch64-gnu.cuda-${CU_UPPERBOUND}.tar.gz", ], ) @@ -111,9 +111,9 @@ http_archive( http_archive( name = "tensorrt_win", build_file = "@//third_party/tensorrt/archive:BUILD", - strip_prefix = "TensorRT-10.13.2.6", + strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/zip/TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/zip/TensorRT-10.13.3.9.Windows.win10.cuda-${CU_UPPERBOUND}.zip", ], ) From ca815e27b8af78d508b44af5d9a1b47fc263d972 Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Thu, 25 Sep 2025 14:43:19 -0700 Subject: [PATCH 5/9] test install fmt --- .github/workflows/build_windows.yml | 2 ++ packaging/pre_build_script.sh | 15 ++------------- packaging/pre_build_script_windows.sh | 4 ++-- 3 files changed, 6 insertions(+), 15 deletions(-) diff --git a/.github/workflows/build_windows.yml b/.github/workflows/build_windows.yml index 07f4de9cb6..f35f8d87ad 100644 --- a/.github/workflows/build_windows.yml +++ b/.github/workflows/build_windows.yml @@ -288,6 +288,8 @@ jobs: BUILD_PARAMS: ${{ inputs.wheel-build-params }} run: | source "${BUILD_ENV_FILE}" + conda install -c conda-forge fmt + if [[ ${{ inputs.is-release-wheel }} == true || ${{ inputs.is-release-tarball }} == true ]]; then # release version for upload to pypi # BUILD_VERSION example: 2.4.0+cu121, we don't want the +cu121 part, so remove +cu121 diff --git a/packaging/pre_build_script.sh b/packaging/pre_build_script.sh index cb2b9a1dd9..67b98c6978 100755 --- a/packaging/pre_build_script.sh +++ b/packaging/pre_build_script.sh @@ -59,22 +59,11 @@ fi export TORCH_BUILD_NUMBER=$(python -c "import torch, urllib.parse as ul; print(ul.quote_plus(torch.__version__))") export TORCH_INSTALL_PATH=$(python -c "import torch, os; print(os.path.dirname(torch.__file__))") -if [[ ${TENSORRT_VERSION} != "" ]]; then - # Replace dependencies in the original pyproject.toml with the current TensorRT version. It is used for CI tests of different TensorRT versions. - # For example, if the current testing TensorRT version is 10.7.0, but the pyproject.toml tensorrt>=10.8.0,<10.9.0, then the following sed command - # will replace tensorrt>=10.8.0,<10.9.0 with tensorrt==10.7.0 - sed -i -e "s/tensorrt>=.*,<.*\"/tensorrt>=${TENSORRT_VERSION},<$(echo "${TENSORRT_VERSION}" | awk -F. '{print $1"."$2+1".0"}')\"/g" \ - -e "s/tensorrt-cu12>=.*,<.*\"/tensorrt-cu12>=${TENSORRT_VERSION},<$(echo "${TENSORRT_VERSION}" | awk -F. '{print $1"."$2+1".0"}')\"/g" \ - -e "s/tensorrt-cu12-bindings>=.*,<.*\"/tensorrt-cu12-bindings>=${TENSORRT_VERSION},<$(echo "${TENSORRT_VERSION}" | awk -F. '{print $1"."$2+1".0"}')\"/g" \ - -e "s/tensorrt-cu12-libs>=.*,<.*\"/tensorrt-cu12-libs>=${TENSORRT_VERSION},<$(echo "${TENSORRT_VERSION}" | awk -F. '{print $1"."$2+1".0"}')\"/g" \ - pyproject.toml -fi - # CU_UPPERBOUND eg:13.0 or 12.9 if [[ ${CU_VERSION:2:2} == "13" ]]; then - CU_UPPERBOUND="13.0" + export CU_UPPERBOUND="13.0" else: - CU_UPPERBOUND="12.9" + export CU_UPPERBOUND="12.9" fi cat toolchains/ci_workspaces/MODULE.bazel.tmpl | envsubst > MODULE.bazel diff --git a/packaging/pre_build_script_windows.sh b/packaging/pre_build_script_windows.sh index 696e90c511..9dde506188 100644 --- a/packaging/pre_build_script_windows.sh +++ b/packaging/pre_build_script_windows.sh @@ -29,9 +29,9 @@ export TORCH_INSTALL_PATH="$(python -c "import torch, os; print(os.path.dirname( # CU_UPPERBOUND eg:13.0 or 12.9 if [[ ${CU_VERSION:2:2} == "13" ]]; then - CU_UPPERBOUND="13.0" + export CU_UPPERBOUND="13.0" else: - CU_UPPERBOUND="12.9" + export CU_UPPERBOUND="12.9" fi cat toolchains/ci_workspaces/MODULE.bazel.tmpl | envsubst > MODULE.bazel From c687f0c63cee0fcd08671e3b37c192c8a6396cf6 Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Thu, 25 Sep 2025 16:25:16 -0700 Subject: [PATCH 6/9] change tensorrt tar name across cu version --- packaging/pre_build_script.sh | 4 +++- packaging/pre_build_script_windows.sh | 4 +++- toolchains/ci_workspaces/MODULE.bazel.tmpl | 2 +- 3 files changed, 7 insertions(+), 3 deletions(-) diff --git a/packaging/pre_build_script.sh b/packaging/pre_build_script.sh index 67b98c6978..2bc0fdc0f1 100755 --- a/packaging/pre_build_script.sh +++ b/packaging/pre_build_script.sh @@ -60,9 +60,11 @@ export TORCH_BUILD_NUMBER=$(python -c "import torch, urllib.parse as ul; print(u export TORCH_INSTALL_PATH=$(python -c "import torch, os; print(os.path.dirname(torch.__file__))") # CU_UPPERBOUND eg:13.0 or 12.9 +# tensorrt tar for linux and windows are different across cuda version +# for sbsa it is the same tar across cuda version if [[ ${CU_VERSION:2:2} == "13" ]]; then export CU_UPPERBOUND="13.0" -else: +else export CU_UPPERBOUND="12.9" fi diff --git a/packaging/pre_build_script_windows.sh b/packaging/pre_build_script_windows.sh index 9dde506188..c1475040db 100644 --- a/packaging/pre_build_script_windows.sh +++ b/packaging/pre_build_script_windows.sh @@ -28,9 +28,11 @@ export CUDA_HOME="$(echo ${CUDA_PATH} | sed -e 's#\\#\/#g')" export TORCH_INSTALL_PATH="$(python -c "import torch, os; print(os.path.dirname(torch.__file__))" | sed -e 's#\\#\/#g')" # CU_UPPERBOUND eg:13.0 or 12.9 +# tensorrt tar for linux and windows are different across cuda version +# for sbsa it is the same tar across cuda version if [[ ${CU_VERSION:2:2} == "13" ]]; then export CU_UPPERBOUND="13.0" -else: +else export CU_UPPERBOUND="12.9" fi cat toolchains/ci_workspaces/MODULE.bazel.tmpl | envsubst > MODULE.bazel diff --git a/toolchains/ci_workspaces/MODULE.bazel.tmpl b/toolchains/ci_workspaces/MODULE.bazel.tmpl index 0a68df8370..e8066e97e6 100644 --- a/toolchains/ci_workspaces/MODULE.bazel.tmpl +++ b/toolchains/ci_workspaces/MODULE.bazel.tmpl @@ -95,7 +95,7 @@ http_archive( build_file = "@//third_party/tensorrt/archive:BUILD", strip_prefix = "TensorRT-10.13.3.9", urls = [ - "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.aarch64-gnu.cuda-${CU_UPPERBOUND}.tar.gz", + "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.aarch64-gnu.cuda-13.0.tar.gz", ], ) From 685632e9fdb11ba4b4c67a29c01f6be1f69a61b9 Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Thu, 25 Sep 2025 18:25:58 -0700 Subject: [PATCH 7/9] move llm test to a seperate folder --- .github/workflows/build-test-linux-x86_64.yml | 1 + .github/workflows/build-test-linux-x86_64_rtx.yml | 1 + .github/workflows/build-test-windows.yml | 1 + .github/workflows/build-test-windows_rtx.yml | 1 + tests/py/dynamo/{models => llm}/test_llm_models.py | 0 5 files changed, 4 insertions(+) rename tests/py/dynamo/{models => llm}/test_llm_models.py (100%) diff --git a/.github/workflows/build-test-linux-x86_64.yml b/.github/workflows/build-test-linux-x86_64.yml index b1630c03be..6d94546177 100644 --- a/.github/workflows/build-test-linux-x86_64.yml +++ b/.github/workflows/build-test-linux-x86_64.yml @@ -177,6 +177,7 @@ jobs: cd tests/py cd dynamo python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_export.xml --ir dynamo models/ + python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_llm.xml llm/ popd tests-py-dynamo-serde: diff --git a/.github/workflows/build-test-linux-x86_64_rtx.yml b/.github/workflows/build-test-linux-x86_64_rtx.yml index 6f04dcdf27..34f9d00568 100644 --- a/.github/workflows/build-test-linux-x86_64_rtx.yml +++ b/.github/workflows/build-test-linux-x86_64_rtx.yml @@ -141,6 +141,7 @@ jobs: cd tests/py cd dynamo python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_export.xml --ir dynamo models/ + python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_llm.xml llm/ popd tests-py-dynamo-serde: diff --git a/.github/workflows/build-test-windows.yml b/.github/workflows/build-test-windows.yml index b97601a0a2..c62515cec4 100644 --- a/.github/workflows/build-test-windows.yml +++ b/.github/workflows/build-test-windows.yml @@ -174,6 +174,7 @@ jobs: cd tests/py cd dynamo python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_export.xml --ir dynamo models/ + python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_llm.xml llm/ popd tests-py-dynamo-serde: diff --git a/.github/workflows/build-test-windows_rtx.yml b/.github/workflows/build-test-windows_rtx.yml index 332db19dc5..9ee768b964 100644 --- a/.github/workflows/build-test-windows_rtx.yml +++ b/.github/workflows/build-test-windows_rtx.yml @@ -143,6 +143,7 @@ jobs: cd tests/py cd dynamo python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_export.xml --ir dynamo models/ + python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/dyn_models_llm.xml llm/ popd tests-py-dynamo-serde: diff --git a/tests/py/dynamo/models/test_llm_models.py b/tests/py/dynamo/llm/test_llm_models.py similarity index 100% rename from tests/py/dynamo/models/test_llm_models.py rename to tests/py/dynamo/llm/test_llm_models.py From 09ff464aa303a4fa4b3b365d189231a0f584b0fd Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Fri, 26 Sep 2025 10:27:03 -0700 Subject: [PATCH 8/9] upgrade windows driver to 580.88 --- packaging/driver_upgrade.bat | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/packaging/driver_upgrade.bat b/packaging/driver_upgrade.bat index 551aa9c7a8..2c173aed81 100644 --- a/packaging/driver_upgrade.bat +++ b/packaging/driver_upgrade.bat @@ -1,9 +1,9 @@ -set WIN_DRIVER_VN=528.89 -set "DRIVER_DOWNLOAD_LINK=https://ossci-windows.s3.amazonaws.com/%WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe" -curl --retry 3 -kL %DRIVER_DOWNLOAD_LINK% --output %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe +set WIN_DRIVER_VN=580.88 +set "DRIVER_DOWNLOAD_LINK=https://ossci-windows.s3.amazonaws.com/%WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe" & REM @lint-ignore +curl --retry 3 -kL %DRIVER_DOWNLOAD_LINK% --output %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe if errorlevel 1 exit /b 1 -start /wait %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe -s -noreboot +start /wait %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe -s -noreboot if errorlevel 1 exit /b 1 -del %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe || ver > NUL +del %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe || ver > NUL From 50a5321b9da31767a2763a3caa6ff41323ef09f3 Mon Sep 17 00:00:00 2001 From: lanluo-nvidia Date: Fri, 26 Sep 2025 11:37:10 -0700 Subject: [PATCH 9/9] skip llm test for rtx on bfloat16 --- .github/workflows/build_windows.yml | 1 - tests/py/dynamo/llm/test_llm_models.py | 2 ++ tests/py/dynamo/models/test_models.py | 2 +- 3 files changed, 3 insertions(+), 2 deletions(-) diff --git a/.github/workflows/build_windows.yml b/.github/workflows/build_windows.yml index f35f8d87ad..41dfd055b4 100644 --- a/.github/workflows/build_windows.yml +++ b/.github/workflows/build_windows.yml @@ -288,7 +288,6 @@ jobs: BUILD_PARAMS: ${{ inputs.wheel-build-params }} run: | source "${BUILD_ENV_FILE}" - conda install -c conda-forge fmt if [[ ${{ inputs.is-release-wheel }} == true || ${{ inputs.is-release-tarball }} == true ]]; then # release version for upload to pypi diff --git a/tests/py/dynamo/llm/test_llm_models.py b/tests/py/dynamo/llm/test_llm_models.py index c1195f247e..31e3efec14 100644 --- a/tests/py/dynamo/llm/test_llm_models.py +++ b/tests/py/dynamo/llm/test_llm_models.py @@ -16,6 +16,8 @@ @pytest.mark.unit @pytest.mark.parametrize("precision", ["FP16", "BF16", "FP32"]) def test_llm_decoder_layer(precision): + if torch_tensorrt.ENABLED_FEATURES.tensorrt_rtx and precision == "BF16": + pytest.skip("TensorRT-RTX does not support bfloat16, skipping test") with torch.inference_mode(): args = argparse.Namespace() diff --git a/tests/py/dynamo/models/test_models.py b/tests/py/dynamo/models/test_models.py index c52b732c42..3d5e2190d2 100644 --- a/tests/py/dynamo/models/test_models.py +++ b/tests/py/dynamo/models/test_models.py @@ -182,7 +182,7 @@ def test_mobilenet_v2(ir, dtype): ) def test_efficientnet_b0(ir, dtype): if torchtrt.ENABLED_FEATURES.tensorrt_rtx and dtype == torch.bfloat16: - pytest.skip("TensorRT-RTX does not support bfloat16") + pytest.skip("TensorRT-RTX does not support bfloat16, skipping test") model = ( timm.create_model("efficientnet_b0", pretrained=True)