[UX] Move Dockerfile DeepGEMM install to tools/install_deepgemm.sh (#23360)

Signed-off-by: mgoin <mgoin64@gmail.com>

docker/Dockerfile

@@ -432,31 +432,11 @@ RUN --mount=type=cache,target=/root/.cache/uv \
         --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
 
 # Install DeepGEMM from source
-ARG DEEPGEMM_GIT_REPO="https://github.com/deepseek-ai/DeepGEMM.git"
 ARG DEEPGEMM_GIT_REF="7b6b5563b9d4c1ae07ffbce7f78ad3ac9204827c"
-RUN --mount=type=cache,target=/root/.cache/uv bash - <<'BASH'
-  . /etc/environment
-    CUDA_MAJOR="${CUDA_VERSION%%.*}"
-    CUDA_MINOR="${CUDA_VERSION#${CUDA_MAJOR}.}"
-    CUDA_MINOR="${CUDA_MINOR%%.*}"
-    if [ "$CUDA_MAJOR" -ge 12 ] && [ "$CUDA_MINOR" -ge 8 ]; then
-        git clone --recursive --shallow-submodules \
-            ${DEEPGEMM_GIT_REPO} deepgemm
-        echo "🏗️  Building DeepGEMM"
-        pushd deepgemm
-            git checkout ${DEEPGEMM_GIT_REF}
-            # Build DeepGEMM
-            # (Based on https://github.com/deepseek-ai/DeepGEMM/blob/main/install.sh)
-            rm -rf build dist
-            rm -rf *.egg-info
-            python3 setup.py bdist_wheel
-            uv pip install --system dist/*.whl
-        popd
-        rm -rf deepgemm
-    else
-        echo "Skipping DeepGEMM installation (requires CUDA 12.8+ but got ${CUDA_VERSION})"
-    fi
-BASH
+COPY tools/install_deepgemm.sh /tmp/install_deepgemm.sh
+RUN --mount=type=cache,target=/root/.cache/uv \
+    VLLM_DOCKER_BUILD_CONTEXT=1 /tmp/install_deepgemm.sh --cuda-version "${CUDA_VERSION}" --ref "${DEEPGEMM_GIT_REF}" \
+    && rm /tmp/install_deepgemm.sh
 
 # Install EP kernels(pplx-kernels and DeepEP), NixL
 COPY tools/ep_kernels/install_python_libraries.sh install_python_libraries.sh