Add exact normalization and rotary CUDA operator modes

.github/workflows/labeler.yml

@@ -15,7 +15,7 @@ jobs:
         ]
     timeout-minutes: 120
     steps:
-    - uses: github/issue-labeler@c1b0f9f52a63158c4adc09425e858e87b32e9685 # v3.4
+    - uses: github/issue-labeler@ccca1e16fcb66762200e616fc29b57b940c1cb67 # v3.5
       with:
         repo-token: "${{ secrets.GITHUB_TOKEN }}"
         configuration-path: .github/labeler.yml

.github/workflows/linux_ci.yml

@@ -2,10 +2,11 @@
 # It leverages a reusable workflow (`reusable_linux_build.yml`) to handle the core build and test logic
 # within Docker containers, ensuring a consistent environment.
 #
-# The workflow consists of five parallel jobs targeting different combinations:
+# The workflow consists of parallel jobs targeting different combinations:
 #   - build-linux-x64-debug: Builds/tests Debug config on Linux x64 (AMD CPU pool), enables AddressSanitizer.
 #   - build-linux-x64-release: Builds/tests Release config on Linux x64 (AMD CPU pool), includes wheel/nuget/benchmark flags.
 #   - orttraining-linux-ci-pipeline: Builds/tests Release config with Training enabled (--enable_training) on Linux x64 (AMD CPU pool).
+#   - build-linux-x64-no-exceptions: Builds MinSizeRel config with exceptions disabled (--disable_exceptions), without unit tests, on Linux x64 (AMD CPU pool).
 #   - build-linux-arm64-debug: Builds/tests Debug config on Linux arm64 (ARM CPU pool); ASan is disabled due to excessive runtime. Includes wheel build.
 #   - build-linux-arm64-release: Builds/tests Release config on Linux arm64 (ARM CPU pool), includes wheel/benchmark flags.
 #
@@ -79,7 +80,7 @@ jobs:
       dockerfile_path: tools/ci_build/github/linux/docker/Dockerfile.manylinux2_28_cpu
       docker_image_repo: onnxruntimecpubuildpythonx64
       # MATERIALIZE=ON needs no explicit pre-build pip install on this leg: the manylinux cp314
-      # image already ships onnx==1.22.0 + numpy==2.4.2 (the exact gate pins) via its
+      # image already ships the pinned onnx + numpy versions via its
       # requirements.txt (install_deps.sh installs it into cp311-cp314). Intentional, not an
       # omission -- unlike the QNN/android legs which run on bare agents and pip-install the pins.
       extra_build_flags: '--use_binskim_compliant_compile_flags --build_wheel --build_nuget --enable_transformers_tool_test --cmake_extra_defines onnxruntime_BUILD_BENCHMARKS=ON --cmake_extra_defines onnxruntime_MATERIALIZE_ONNX_NODE_TESTS=ON'
@@ -103,6 +104,22 @@ jobs:
     secrets:
       GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
+  build-linux-x64-no-exceptions:
+    name: Build Linux x64 MinSizeRel (Exceptions Disabled)
+    uses: ./.github/workflows/reusable_linux_build.yml
+    with:
+      pool_name: "onnxruntime-github-Ubuntu2204-AMD-CPU"
+      build_config: MinSizeRel
+      architecture: x64
+      dockerfile_path: tools/ci_build/github/linux/docker/inference/x86_64/default/cpu/Dockerfile
+      docker_image_repo: onnxruntimecpubuildcix64
+      # TODO: fix the unit test code (unguarded try/catch and gsl::narrow) to build with exceptions disabled, then build and run the tests here.
+      extra_build_flags: '--use_binskim_compliant_compile_flags --disable_exceptions --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF'
+      run_tests: false
+      job_identifier: build-linux-x64-no-exceptions
+    secrets:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
   # --- arm64 Builds ---
   build-linux-arm64-debug:
     name: Build Linux arm64 Debug

.github/workflows/linux_cuda_ci.yml

@@ -127,6 +127,7 @@ jobs:
           mode: 'test' # Set mode to test
           execution_providers: 'cuda'
           extra_build_flags: >-
+            --enable_pybind
             --use_binskim_compliant_compile_flags
             --cuda_version=13.0
             --cuda_home=/usr/local/cuda-13.0
@@ -137,4 +138,4 @@ jobs:
             --cmake_extra_defines onnxruntime_QUICK_BUILD=ON
             --cmake_extra_defines onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS=ON
             --cmake_extra_defines onnxruntime_BUILD_CUDA_QUANT_PREPROCESS=ON
-          python_path_prefix: 'PATH=/opt/python/cp310-cp310/bin:$PATH'
+          python_path_prefix: 'export ORT_RUN_CUDA_TACTIC_CACHE_TESTS=1 PATH=/opt/python/cp310-cp310/bin:$PATH'

.github/workflows/linux_cuda_plugin_ci.yml

@@ -136,6 +136,8 @@ jobs:
             -v ${{ runner.temp }}/Release:/build/Release \
             -e NVIDIA_VISIBLE_DEVICES=all \
             -e ORT_TEST_CUDA_PLUGIN_EP=1 \
+            -e ORT_RUN_CUDA_TACTIC_CACHE_TESTS=1 \
+            -e ORT_CUDA_PLUGIN_PATH=/build/Release/Release/libonnxruntime_providers_cuda.so \
             ${{ steps.build_docker_image_step.outputs.full-image-name }} \
             bash -c "
               set -ex
@@ -144,6 +146,8 @@ jobs:
               python -m pip install /build/Release/Release/dist/onnxruntime*.whl
               python -m pip install numpy onnx
               python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
+              cd /build/Release/Release
+              python /onnxruntime_src/onnxruntime/test/python/onnxruntime_test_python_fpa_intb_tune.py
               cd /onnxruntime_src/onnxruntime/test/python/transformers
               python test_cuda_plugin_ep.py
             "

.github/workflows/title-only-labeler.yml

@@ -15,7 +15,7 @@ jobs:
         ]
     timeout-minutes: 120
     steps:
-    - uses: github/issue-labeler@c1b0f9f52a63158c4adc09425e858e87b32e9685 # v3.4
+    - uses: github/issue-labeler@ccca1e16fcb66762200e616fc29b57b940c1cb67 # v3.5
       with:
         repo-token: "${{ secrets.GITHUB_TOKEN }}"
         configuration-path: .github/title-only-labeler.yml

ThirdPartyNotices.txt

@@ -19,76 +19,6 @@ General Public License.
 
 _____
 
-Intel Math Kernel Library (Intel MKL)
-
-Intel Simplified Software License (Version April 2018)
-
-Copyright (c) 2018 Intel Corporation.
-
-Use and Redistribution.  You may use and redistribute the software (the “Software”), without modification,
-provided the following conditions are met:
-
-* Redistributions must reproduce the above copyright notice and the following terms of use in the Software
-and in the documentation and/or other materials provided with the distribution.
-
-* Neither the name of Intel nor the names of its suppliers may be used to endorse or promote products
-derived from this Software without specific prior written permission.
-
-* No reverse engineering, decompilation, or disassembly of this Software is permitted.
-
-Limited patent license.  Intel grants you a world-wide, royalty-free, non-exclusive license under patents it now
-or hereafter owns or controls to make, have made, use, import, offer to sell and sell (“Utilize”) this Software,
-but solely to the extent that any such patent is necessary to Utilize the Software alone. The patent license
-shall not apply to any combinations which include this software.  No hardware per se is licensed hereunder.
-
-Third party and other Intel programs.  “Third Party Programs” are the files listed in the “third-party-programs.txt”
-text file that is included with the Software and may include Intel programs under separate license terms.
-Third Party Programs, even if included with the distribution of the Materials, are governed by
-separate license terms and those license terms solely govern your use of those programs.
-
-DISCLAIMER.  THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
-INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS
-FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT ARE DISCLAIMED. THIS SOFTWARE IS
-NOT INTENDED FOR USE IN SYSTEMS OR APPLICATIONS WHERE FAILURE OF THE SOFTWARE
-MAY CAUSE PERSONAL INJURY OR DEATH AND YOU AGREE THAT YOU ARE FULLY RESPONSIBLE FOR ANY
-CLAIMS, COSTS, DAMAGES, EXPENSES, AND ATTORNEYS’ FEES ARISING OUT OF ANY SUCH USE,
-EVEN IF ANY CLAIM ALLEGES THAT INTEL WAS NEGLIGENT REGARDING THE DESIGN OR MANUFACTURE OF
-THE MATERIALS.
-
-LIMITATION OF LIABILITY. IN NO EVENT WILL INTEL BE LIABLE FOR ANY DIRECT, INDIRECT,
-INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
-WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
-ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY
-OF SUCH DAMAGE. YOU AGREE TO INDEMNIFY AND HOLD INTEL HARMLESS AGAINST ANY CLAIMS
-AND EXPENSES RESULTING FROM YOUR USE OR UNAUTHORIZED USE OF THE SOFTWARE.
-
-No support.  Intel may make changes to the Software, at any time without notice, and is not obligated to
-support, update or provide training for the Software.
-
-Termination. Intel may terminate your right to use the Software in the event of your breach of this Agreement
-and you fail to cure the breach within a reasonable period of time.
-
-Feedback.  Should you provide Intel with comments, modifications, corrections, enhancements or other input
-(“Feedback”) related to the Software Intel will be free to use, disclose, reproduce, license or otherwise
-distribute or exploit the Feedback in its sole discretion without any obligations or restrictions of any kind,
-including without limitation, intellectual property rights or licensing obligations.
-
-Compliance with laws.  You agree to comply with all relevant laws and regulations governing your use,
-transfer, import or export (or prohibition thereof) of the Software.
-
-Governing law.  All disputes will be governed by the laws of the United States of America and the State of
-Delaware without reference to conflict of law principles and subject to the exclusive jurisdiction of the state or
-federal courts sitting in the State of Delaware, and each party agrees that it submits to the personal
-jurisdiction and venue of those courts and waives any objections. The United Nations Convention on
-Contracts for the International Sale of Goods (1980) is specifically excluded and will not apply to the
-Software.
-
-*Other names and brands may be claimed as the property of others.
-
-_____
-
 protocolbuffers/protobuf
 
 Copyright 2008 Google Inc.  All rights reserved.