[CI] Update Windows CI (#3322)

This PR updates the Windows CI with the latest one-line build flow.

Author: MasterJH5574

ci/jenkinsfile.groovy

@@ -125,7 +125,6 @@ stage('Build') {
           sh(script: "ls -alh", label: 'Show work directory')
           sh(script: "${pkg_cuda} conda env export --name py312", label: 'Checkout version')
           sh(script: "${pkg_cuda} -j 8 -v \$HOME/.ccache /ccache conda run -n py312 ./ci/task/build_lib.sh", label: 'Build MLC LLM runtime')
-          sh(script: "${pkg_cuda} -j 8 conda run -n py312 ./ci/task/build_wheel.sh", label: 'Build MLC LLM wheel')
           sh(script: "${pkg_cuda} -j 1 conda run -n py312 ./ci/task/build_clean.sh", label: 'Clean up after build')
           sh(script: "ls -alh ./wheels/", label: 'Build artifact')
           pack_lib('mlc_wheel_cuda', 'wheels/*.whl')
@@ -139,7 +138,6 @@ stage('Build') {
     //       sh(script: "ls -alh", label: 'Show work directory')
     //       sh(script: "${pkg_rocm} conda env export --name py38", label: 'Checkout version')
     //       sh(script: "${pkg_rocm} -j 8 conda run -n py38 ./ci/task/build_lib.sh", label: 'Build MLC LLM runtime')
-    //       sh(script: "${pkg_rocm} -j 8 conda run -n py38 ./ci/task/build_wheel.sh", label: 'Build MLC LLM wheel')
     //       sh(script: "${pkg_rocm} -j 1 conda run -n py38 ./ci/task/build_clean.sh", label: 'Clean up after build')
     //       sh(script: "ls -alh ./wheels/", label: 'Build artifact')
     //       pack_lib('mlc_wheel_rocm', 'wheels/*.whl')
@@ -153,7 +151,6 @@ stage('Build') {
           sh(script: "ls -alh", label: 'Show work directory')
           sh(script: "conda env export --name mlc-llm-ci", label: 'Checkout version')
           sh(script: "NUM_THREADS=6 GPU=metal conda run -n mlc-llm-ci ./ci/task/build_lib.sh", label: 'Build MLC LLM runtime')
-          sh(script: "NUM_THREADS=6 GPU=metal conda run -n mlc-llm-ci ./ci/task/build_wheel.sh", label: 'Build MLC LLM wheel')
           sh(script: "NUM_THREADS=6 GPU=metal conda run -n mlc-llm-ci ./ci/task/build_clean.sh", label: 'Clean up after build')
           sh(script: "ls -alh ./wheels/", label: 'Build artifact')
           pack_lib('mlc_wheel_metal', 'wheels/*.whl')
@@ -167,7 +164,6 @@ stage('Build') {
           sh(script: "ls -alh", label: 'Show work directory')
           sh(script: "${pkg_cpu} conda env export --name py312", label: 'Checkout version')
           sh(script: "${pkg_cpu} -j 8 conda run -n py312 ./ci/task/build_lib.sh", label: 'Build MLC LLM runtime')
-          sh(script: "${pkg_cpu} -j 8 conda run -n py312 ./ci/task/build_wheel.sh", label: 'Build MLC LLM wheel')
           sh(script: "${pkg_cpu} -j 1 conda run -n py312 ./ci/task/build_clean.sh", label: 'Clean up after build')
           sh(script: "ls -alh ./wheels/", label: 'Build artifact')
           pack_lib('mlc_wheel_vulkan', 'wheels/*.whl')

ci/task/build_lib.sh

@@ -20,7 +20,7 @@ if [[ ${GPU} != metal ]]; then
     source /opt/rh/gcc-toolset-11/enable # GCC-11 is the hightest GCC version compatible with NVCC < 12
 fi
 
-mkdir -p $WORKSPACE_CWD/build/ && cd $WORKSPACE_CWD/build/
+mkdir -p $WORKSPACE_CWD/build
 if [[ ${GPU} == rocm* ]]; then
     echo set\(USE_VULKAN ON\) >>config.cmake
     echo set\(USE_ROCM ON\) >>config.cmake
@@ -33,7 +33,6 @@ elif [[ ${GPU} == cuda* ]]; then
     echo set\(USE_CUDA ON\) >>config.cmake
     echo set\(USE_CUBLAS ON\) >>config.cmake
     echo set\(USE_NCCL ON\) >>config.cmake
-    echo set\(USE_CUTLASS ON\) >>config.cmake
 elif [[ ${GPU} == metal ]]; then
     export CCACHE_DIR=$HOME/ci/ccache
     echo set\(USE_METAL ON\) >>config.cmake
@@ -42,3 +41,27 @@ else
 fi
 
 cat config.cmake
+
+AUDITWHEEL_OPTS="--plat ${AUDITWHEEL_PLAT} -w repaired_wheels/"
+AUDITWHEEL_OPTS="--exclude libtvm --exclude libtvm_runtime --exclude libtvm_ffi --exclude libvulkan ${AUDITWHEEL_OPTS}"
+if [[ ${GPU} == rocm* ]]; then
+    AUDITWHEEL_OPTS="--exclude libamdhip64 --exclude libhsa-runtime64 --exclude librocm_smi64 --exclude librccl ${AUDITWHEEL_OPTS}"
+elif [[ ${GPU} == cuda* ]]; then
+    AUDITWHEEL_OPTS="--exclude libcuda --exclude libcudart --exclude libnvrtc --exclude libcublas --exclude libcublasLt ${AUDITWHEEL_OPTS}"
+fi
+
+rm -rf ${WORKSPACE_CWD}/dist
+cd ${WORKSPACE_CWD} && pip wheel --no-deps -w dist . -v
+
+rm -rf ${WORKSPACE_CWD}/wheels/
+if [[ ${GPU} != metal ]]; then
+    mkdir -p ${WORKSPACE_CWD}/repaired_wheels
+    rm -rf ${WORKSPACE_CWD}/repaired_wheels/*
+    auditwheel repair ${AUDITWHEEL_OPTS} dist/*.whl
+    mv ${WORKSPACE_CWD}/repaired_wheels/ ${WORKSPACE_CWD}/wheels/
+else
+    mkdir ${WORKSPACE_CWD}/wheels/
+    mv dist/*.whl ${WORKSPACE_CWD}/wheels/
+fi
+
+chown -R $ENV_USER_ID:$ENV_GROUP_ID ${WORKSPACE_CWD}/wheels/

ci/task/build_wheel.sh

@@ -1,15 +1,9 @@
 cd mlc-llm
 rd /s /q build
 mkdir build
-cd build
 
-cmake -A x64 -Thost=x64 ^
-      -G "Visual Studio 17 2022" ^
-      -DUSE_VULKAN=ON ^
-      ..
+echo set(USE_VULKAN ON) >> config.cmake
 
-if %errorlevel% neq 0 exit %errorlevel%
-
-cmake --build . --parallel 3 --config Release -- /m
+pip install . -v
 
 if %errorlevel% neq 0 exit %errorlevel%