diff --git a/.github/workflows/win-cuda-x64-build.yml b/.github/workflows/win-cuda-x64-build.yml index 45543c8b70..d799950d1c 100644 --- a/.github/workflows/win-cuda-x64-build.yml +++ b/.github/workflows/win-cuda-x64-build.yml @@ -105,7 +105,11 @@ jobs: - name: Configure CMake run: | $env:SCCACHE_DIR = "${{ github.workspace }}\sccache" + # Limit CUDA archs to SM 86 (matches the A10 runner GPU) to keep CI build + # time and memory usage manageable. The full SM list in CMakePresets.json + # is intended for release packaging, not CI validation. cmake --preset windows_x64_cuda_release -T cuda=${{ env.cuda_dir }}\\v${{ env.cuda_version }} -DTEST_PHI2=True -DTEST_QWEN_2_5=True -DENABLE_JAVA=ON -DUSE_GUIDANCE=ON ` + -DCMAKE_CUDA_ARCHITECTURES=86-real ` -DCMAKE_C_COMPILER_LAUNCHER=sccache ` -DCMAKE_CXX_COMPILER_LAUNCHER=sccache diff --git a/.github/workflows/win-winml-x64-build.yml b/.github/workflows/win-winml-x64-build.yml index 9ef669a634..db01dee5da 100644 --- a/.github/workflows/win-winml-x64-build.yml +++ b/.github/workflows/win-winml-x64-build.yml @@ -62,7 +62,10 @@ jobs: - name: Configure CMake run: | - cmake --preset windows_x64_winml_relwithdebinfo -T cuda=${{ env.cuda_dir }}\\v${{ env.cuda_version }} -DWINML_SDK_VERSION='1.8.2091' -DENABLE_JAVA=ON -DUSE_GUIDANCE=ON + # Limit CUDA archs to SM 86 (matches the A10 runner GPU) to keep CI build + # time and memory usage manageable. The preset's default arch list is for + # release packaging, not CI validation. + cmake --preset windows_x64_winml_relwithdebinfo -T cuda=${{ env.cuda_dir }}\\v${{ env.cuda_version }} -DWINML_SDK_VERSION='1.8.2091' -DENABLE_JAVA=ON -DUSE_GUIDANCE=ON -DCMAKE_CUDA_ARCHITECTURES=86-real - name: Build with CMake run: | diff --git a/test/python/requirements.txt b/test/python/requirements.txt index e0bc731117..2af195bc50 100644 --- a/test/python/requirements.txt +++ b/test/python/requirements.txt @@ -8,4 +8,3 @@ pytest onnx onnx_ir>=0.1.3 transformers<5.0.0 -huggingface_hub[cli,hf_xet] \ No newline at end of file