diff --git a/.github/workflows/build-gpu-k8s-offline-package.yml b/.github/workflows/build-gpu-k8s-offline-package.yml
new file mode 100644
index 0000000..3a72c9c
--- /dev/null
+++ b/.github/workflows/build-gpu-k8s-offline-package.yml
@@ -0,0 +1,22 @@
+name: Build GPU K8s offline package
+
+on:
+  workflow_dispatch:
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: Build offline package
+        run: |
+          bash scripts/create-gpu-k8s-offline-package.sh
+
+      - name: Upload artifact
+        uses: actions/upload-artifact@v3
+        with:
+          name: gpu_k8s_offline_packages
+          path: gpu_k8s_offline_packages.tar.gz
+
diff --git a/scripts/create-gpu-k8s-offline-package.sh b/scripts/create-gpu-k8s-offline-package.sh
new file mode 100755
index 0000000..5994f8b
--- /dev/null
+++ b/scripts/create-gpu-k8s-offline-package.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+set -euo pipefail
+
+# Versions
+K8S_VERSION="${K8S_VERSION:-labring/kubernetes:v1.25.16}"
+CILIUM_VERSION="${CILIUM_VERSION:-labring/cilium:v1.13.4}"
+HELM_VERSION="${HELM_VERSION:-labring/helm:v3.9.4}"
+NERDCTL_VERSION="${NERDCTL_VERSION:-2.1.2}"
+NVIDIA_PLUGIN_VERSION="${NVIDIA_PLUGIN_VERSION:-v0.17.1}"
+NVIDIA_DRIVER_VERSION="${NVIDIA_DRIVER_VERSION:-nvidia-driver-535}"
+CUDA_SAMPLE_IMAGE="${CUDA_SAMPLE_IMAGE:-nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0}"
+
+IMAGES=(
+  "$K8S_VERSION"
+  "$CILIUM_VERSION"
+  "$HELM_VERSION"
+  "nvcr.io/nvidia/k8s-device-plugin:${NVIDIA_PLUGIN_VERSION}"
+  "$CUDA_SAMPLE_IMAGE"
+)
+
+WORKDIR="offline"
+mkdir -p "$WORKDIR/images" "$WORKDIR/packages"
+
+# Download required APT packages
+APT_PACKAGES=(
+  curl gnupg2 ca-certificates lsb-release apt-transport-https \
+  software-properties-common openssh-client openssh-server uidmap \
+  containerd "$NVIDIA_DRIVER_VERSION" nvidia-container-toolkit
+)
+
+sudo apt-get update -y
+sudo apt-get install --download-only -y "${APT_PACKAGES[@]}"
+cp /var/cache/apt/archives/*.deb "$WORKDIR/packages/"
+sudo apt-get clean
+
+# Download sealos deb
+curl -L -o "$WORKDIR/sealos_5.0.1_linux_amd64.deb" \
+  https://github.com/labring/sealos/releases/download/v5.0.1/sealos_5.0.1_linux_amd64.deb
+
+# Download nerdctl archive
+nerdctl_archive="nerdctl-full-${NERDCTL_VERSION}-linux-amd64.tar.gz"
+curl -L -o "$WORKDIR/${nerdctl_archive}" \
+  "https://github.com/containerd/nerdctl/releases/download/v${NERDCTL_VERSION}/${nerdctl_archive}"
+
+# Pull required container images
+for img in "${IMAGES[@]}"; do
+  echo "Pulling $img"
+  docker pull "$img"
+done
+
+docker save -o "$WORKDIR/images/gpu_k8s_images.tar" "${IMAGES[@]}"
+
+# Download NVIDIA device plugin manifest and gpg key
+PLUGIN_FILE="nvidia-device-plugin.yml"
+curl -L -o "$WORKDIR/${PLUGIN_FILE}" \
+  "https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/${NVIDIA_PLUGIN_VERSION}/deployments/static/${PLUGIN_FILE}"
+curl -L -o "$WORKDIR/nvidia-gpgkey" https://nvidia.github.io/nvidia-docker/gpgkey
+
+# Include deployment script
+cp "$(dirname "$0")/gpu-k8s.sh" "$WORKDIR/"
+
+# Create final archive
+TAR_NAME="gpu_k8s_offline_packages.tar.gz"
+tar -czf "$TAR_NAME" -C "$WORKDIR" .
+
+echo "Created $TAR_NAME"
diff --git a/scripts/gpu-k8s.sh b/scripts/gpu-k8s.sh
new file mode 100644
index 0000000..3e1b506
--- /dev/null
+++ b/scripts/gpu-k8s.sh
@@ -0,0 +1,253 @@
+#!/bin/bash
+set -euo pipefail
+
+# === 全局变量 ===
+MASTER_IP=$(hostname -I | awk '{print $1}')
+USER=${USER:-$(whoami)}
+SSH_KEY="${HOME}/.ssh/id_rsa"
+K8S_VERSION="labring/kubernetes:v1.25.16"
+CILIUM_VERSION="labring/cilium:v1.13.4"
+HELM_VERSION="labring/helm:v3.9.4"
+NVIDIA_DRIVER_VERSION="nvidia-driver-535"
+NVIDIA_PLUGIN_VERSION="v0.17.1"
+NERDCTL_VERSION="2.1.2"
+PROXY_ADDR="http://127.0.0.1:1081"
+USE_PROXY=${USE_PROXY:-false}
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+OFFLINE_DIR=${OFFLINE_DIR:-$SCRIPT_DIR}
+
+# === 选项代理 ===
+configure_proxy() {
+  if [ "$USE_PROXY" = true ]; then
+    export http_proxy=$PROXY_ADDR
+    export https_proxy=$PROXY_ADDR
+    export HTTP_PROXY=$PROXY_ADDR
+    export HTTPS_PROXY=$PROXY_ADDR
+    echo "🌐 代理已启用: $PROXY_ADDR"
+  else
+    unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY
+    echo "🌐 代理已关闭"
+  fi
+}
+
+proxy_curl() {
+  if [ "$USE_PROXY" = true ]; then
+    curl --proxy "$PROXY_ADDR" "$@"
+  else
+    curl "$@"
+  fi
+}
+
+load_offline_images() {
+  local tar="${OFFLINE_DIR}/images/gpu_k8s_images.tar"
+  [ -f "$tar" ] || return 0
+  echo "📦 Loading images from $tar"
+  if command -v nerdctl &>/dev/null; then
+    sudo nerdctl load -i "$tar"
+  elif command -v docker &>/dev/null; then
+    docker load -i "$tar"
+  else
+    echo "❌ 无法找到 nerdctl 或 docker 用于加载镜像"
+    return 1
+  fi
+}
+
+install_all_offline_packages() {
+  if [ -d "${OFFLINE_DIR}/packages" ]; then
+    echo "📦 Using offline deb packages"
+    sudo dpkg -i ${OFFLINE_DIR}/packages/*.deb 2>/dev/null || sudo apt-get -f install -y
+    return 0
+  fi
+  return 1
+}
+
+install_base() {
+  echo "[1/8] 安装基础依赖"
+  install_all_offline_packages || {
+    sudo apt-get update -y
+    sudo apt-get install -y curl gnupg2 ca-certificates lsb-release \
+      apt-transport-https software-properties-common openssh-client \
+      openssh-server uidmap containerd ${NVIDIA_DRIVER_VERSION} nvidia-container-toolkit
+  }
+}
+
+install_containerd() {
+  echo "[2/8] 安装 containerd + nerdctl"
+  sudo apt-get purge -y docker.io docker-ce docker-ce-cli containerd.io || true
+  if ! install_all_offline_packages; then
+    sudo apt-get install -y containerd
+  fi
+
+  archive="nerdctl-full-${NERDCTL_VERSION}-linux-amd64.tar.gz"
+  if [ -f "${OFFLINE_DIR}/${archive}" ]; then
+    tmpdir="$OFFLINE_DIR"
+  else
+    tmpdir=$(mktemp -d)
+    url="https://github.com/containerd/nerdctl/releases/download/v${NERDCTL_VERSION}/${archive}"
+    echo "🔽 下载 nerdctl: $url"
+    proxy_curl -fLo "${tmpdir}/${archive}" "$url"
+  fi
+
+  echo "📅 解压 nerdctl 到 /usr/local"
+  sudo tar -xzf "${tmpdir}/${archive}" -C /usr/local
+
+  sudo mkdir -p /etc/containerd
+  sudo containerd config default | sudo tee /etc/containerd/config.toml > /dev/null
+  sudo systemctl enable --now containerd
+  nerdctl --version && echo "✅ nerdctl 安装成功" || echo "❌ nerdctl 安装失败"
+}
+
+install_nvidia() {
+  echo "[3/8] 安装 NVIDIA 驱动和容器工具"
+  distribution="ubuntu22.04"
+  if [ -f "${OFFLINE_DIR}/nvidia-gpgkey" ]; then
+    sudo install -m 0644 "${OFFLINE_DIR}/nvidia-gpgkey" /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+  else
+    proxy_curl -sL https://nvidia.github.io/nvidia-docker/gpgkey | \
+      sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+  fi
+
+  if ! install_all_offline_packages; then
+    proxy_curl -sL https://nvidia.github.io/nvidia-docker/${distribution}/nvidia-docker.list | \
+      sed 's|^deb |deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] |' | \
+      sudo tee /etc/apt/sources.list.d/nvidia-docker.list
+    sudo apt-get update -y
+    sudo apt-get install -y ${NVIDIA_DRIVER_VERSION} nvidia-container-toolkit
+  fi
+  sudo nvidia-ctk runtime configure --runtime=containerd --set-as-default
+  sudo systemctl restart containerd
+  if ! command -v nvidia-smi >/dev/null; then echo "❌ nvidia-smi 未找到"; exit 1; fi
+  nvidia-smi || { echo "❌ NVIDIA 驱动有问题"; exit 1; }
+}
+
+install_sealos() {
+  echo "[4/8] 安装 Sealos"
+  if command -v sealos &>/dev/null; then
+    return
+  fi
+  if [ -f "${OFFLINE_DIR}/sealos_5.0.1_linux_amd64.deb" ]; then
+    sudo dpkg -i "${OFFLINE_DIR}/sealos_5.0.1_linux_amd64.deb"
+  else
+    proxy_curl -sfL https://raw.githubusercontent.com/labring/sealos/main/scripts/install.sh | bash
+  fi
+}
+
+setup_ssh() {
+  echo "[5/8] 配置 SSH 免密"
+  [ ! -f "${SSH_KEY}" ] && ssh-keygen -f "${SSH_KEY}" -N ""
+  cat "${SSH_KEY}.pub" >> ~/.ssh/authorized_keys
+  chmod 600 ~/.ssh/authorized_keys && chmod 700 ~/.ssh
+  sudo systemctl enable --now ssh || sudo systemctl enable --now sshd
+}
+
+deploy_k8s() {
+  echo "[6/8] 使用 Sealos 部署 K8s"
+  load_offline_images || true
+  sealos run "${K8S_VERSION}" "${CILIUM_VERSION}" "${HELM_VERSION}" \
+    --masters "${MASTER_IP}" --user "${USER}" --pk "${SSH_KEY}" \
+    --env '{}' --cmd "kubeadm init --skip-phases=addon/kube-proxy"
+
+  echo "[6.1] 禁用 sealos containerd, 启用系统 containerd"
+  sudo systemctl disable --now sealos-containerd || true
+  sudo systemctl enable --now containerd
+  sleep 3
+  sudo systemctl status containerd --no-pager | grep Active
+}
+
+deploy_plugin() {
+  echo "[7/8] 部署 NVIDIA Device Plugin"
+  local plugin_file="${OFFLINE_DIR}/nvidia-device-plugin.yml"
+  if [ -f "$plugin_file" ]; then
+    kubectl apply -f "$plugin_file"
+  else
+    plugin_url="https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/${NVIDIA_PLUGIN_VERSION}/deployments/static/nvidia-device-plugin.yml"
+    if [ "$USE_PROXY" = true ]; then
+      HTTPS_PROXY=$PROXY_ADDR HTTP_PROXY=$PROXY_ADDR \
+      kubectl apply -f "$plugin_url"
+    else
+      kubectl apply -f "$plugin_url"
+    fi
+  fi
+  sleep 15
+  kubectl -n kube-system get pods | grep nvidia || echo "⚠️ 插件未启动"
+  kubectl describe node | grep -A10 Capacity | grep -i nvidia
+}
+
+run_test() {
+  echo "[8/8] 运行 CUDA vectoradd GPU 测试"
+  kubectl apply -f - <<'YAML'
+apiVersion: v1
+kind: Pod
+metadata:
+  name: gpu-pod
+spec:
+  restartPolicy: Never
+  containers:
+  - name: cuda-test
+    image: nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0
+    resources:
+      limits:
+        nvidia.com/gpu: 1
+  tolerations:
+  - key: nvidia.com/gpu
+    operator: Exists
+    effect: NoSchedule
+YAML
+  kubectl wait pod/gpu-pod --for=condition=Succeeded --timeout=120s || true
+  kubectl logs gpu-pod || echo "⚠️ 未获取日志"
+}
+
+show_help() {
+  echo -e "用法: ./gpu-k8s.sh [阶段参数...]\n"
+  echo "可用阶段:"
+  echo "  --install-base         安装基础依赖"
+  echo "  --install-containerd   安装 containerd + nerdctl"
+  echo "  --install-nvidia       安装 NVIDIA 驱动和工具"
+  echo "  --install-sealos       安装 Sealos"
+  echo "  --setup-ssh            配置 SSH 免密"
+  echo "  --deploy-k8s           使用 Sealos 部署 Kubernetes"
+  echo "  --deploy-plugin        部署 NVIDIA Device Plugin"
+  echo "  --run-test             运行 GPU 测试"
+  echo "  --all                  全部步骤执行"
+  echo ""
+  echo "环境变量:"
+  echo "  OFFLINE_DIR           指定离线包解压目录，默认为脚本所在目录"
+  echo -e "\n示例命令\t\t\t说明"
+  echo "USE_PROXY=true ./gpu-k8s.sh --install-nvidia      # 只安装 NVIDIA 工具包并走代理"
+  echo "USE_PROXY=false ./gpu-k8s.sh --all                # 全流程执行但不使用代理"
+  echo "./gpu-k8s.sh --install-sealos --deploy-k8s        # 默认关闭代理执行指定阶段"
+  echo "OFFLINE_DIR=/path/to/offline ./gpu-k8s.sh --all   # 使用离线包运行"
+}
+
+# === 执行 ===
+configure_proxy
+
+if [ $# -eq 0 ]; then
+  show_help
+  exit 1
+fi
+
+for arg in "$@"; do
+  case "$arg" in
+    --install-base) install_base ;;
+    --install-containerd) install_containerd ;;
+    --install-nvidia) install_nvidia ;;
+    --install-sealos) install_sealos ;;
+    --setup-ssh) setup_ssh ;;
+    --deploy-k8s) deploy_k8s ;;
+    --deploy-plugin) deploy_plugin ;;
+    --run-test) run_test ;;
+    --all)
+      install_base
+      install_containerd
+      install_nvidia
+      install_sealos
+      setup_ssh
+      deploy_k8s
+      deploy_plugin
+      run_test
+      ;;
+    *) show_help; exit 1 ;;
+  esac
+  echo
+done