feat(build): add CUDA toolkit configuration and MMQ support

.github/workflows/build.yml

@@ -394,6 +394,8 @@ jobs:
         uses: Jimver/cuda-toolkit@master
         with:
           cuda: ${{ matrix.cuda_version }}
+          method: network
+          sub-packages: '["nvcc", "cudart"]'
       
       - name: Install Dependencies
         run: |
@@ -406,6 +408,7 @@ jobs:
             -DCMAKE_BUILD_TYPE=Release \
             -DGGML_DIR=ggml \
             -DGGML_CUDA=ON \
+            -DGGML_CUDA_FORCE_MMQ=ON \
             -DCMAKE_CUDA_ARCHITECTURES="75;80;86;89" \
             -DMBR_BUILD_TESTS=OFF \
             -DMBR_BUILD_CLI=ON
@@ -477,6 +480,8 @@ jobs:
         uses: Jimver/cuda-toolkit@master
         with:
           cuda: ${{ matrix.cuda_version }}
+          method: network
+          sub-packages: '["nvcc", "cudart", "visual_studio_integration"]'
         
       - name: Install Ninja
         run: choco install ninja -y
@@ -488,6 +493,7 @@ jobs:
           cmake -B build -G "Ninja Multi-Config" ^
             -DGGML_DIR=ggml ^
             -DGGML_CUDA=ON ^
+            -DGGML_CUDA_FORCE_MMQ=ON ^
             -DCMAKE_CUDA_ARCHITECTURES="75;80;86;89" ^
             -DMBR_BUILD_TESTS=OFF ^
             -DMBR_BUILD_CLI=ON