bool has_gpu0 = false;
  bool has_gpu1 = false;
  for (int i = 0; i < num_devices; ++i) {
    const char* dev = TF_DeviceListName(devices, i, status.get());
    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
    string device_name(dev);
    if (device_name.find("GPU:0") != string::npos) {
      has_gpu0 = true;
    }
    if (device_name.find("GPU:1") != string::npos) {

# CuDNN: https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#ubuntu-network-installation
libcudnn8=8.9.6.50-1+cuda12.2
libcudnn8-dev=8.9.6.50-1+cuda12.2

# This can be removed once NVIDIA publishes a cuda-12.3.2 Docker image.
# For now it ensures that we install at least version 12.3.107 of PTXAS,
# since 12.3.103 has a bug.

set -xe

source ~/.bashrc
VERSION=$1
REQUIREMENTS=$2

add-apt-repository ppa:deadsnakes/ppa
# Install Python packages for this container's version
cat >pythons.txt <<EOF
$VERSION
$VERSION-dev
$VERSION-venv
$VERSION-distutils
EOF
/setup.packages.sh pythons.txt

# Re-link pyconfig.h from aarch64-linux-gnu into the devtoolset directory
# for any Python version present
pushd /usr/include/aarch64-linux-gnu

from tensorflow.python.util.lazy_loader import LazyLoader as _LazyLoader
from tensorflow.python.util.lazy_loader import KerasLazyLoader as _KerasLazyLoader

# API IMPORTS PLACEHOLDER

# WRAPPER_PLACEHOLDER

if "dev" in __version__:   # pylint: disable=undefined-variable
  _logging.warning("""

  TensorFlow's `tf-nightly` package will soon be updated to TensorFlow 2.0.

  Please upgrade your code to TensorFlow 2.0:

  virtual bool IsCustomDevice(const string& device_name) = 0;

  // Register a custom device. It will return error is the device name is
  // already registered.
  // TODO(tfrt-devs): Remove this method. Let caller register it directly into
  // CustomDeviceOpHandler.
  virtual Status RegisterCustomDevice(const string& name,
                                      std::unique_ptr<CustomDevice> device) = 0;

# Note that here we want the latest version that matches TF major.minor version
# Note that we must use nightly here as these are used in nightly jobs
# For release jobs, we will pin these on the release branch
keras-nightly ~= 3.0.0.dev
tb-nightly ~= 2.17.0.a

# Test dependencies
grpcio >= 1.24.3, < 2.0
portpicker == 1.6.0
scipy == 1.11.3
requests == 2.31.0
packaging==23.2
setuptools==68.2.2

# Writes the list of tests that would be run from BAZEL_TEST_COMMAND to OUTPUT.
# Hides all extra output and always exits with success for now.

OUTPUT=$1
shift

  fi
fi

# Keep the existing "tf" container if it's already present.
# The container is not cleaned up automatically! Remove it with:
# docker rm tf
if ! docker container inspect tf >/dev/null 2>&1 ; then
  # Pass all existing TFCI_ variables into the Docker container
  env_file=$(mktemp)
  env | grep ^TFCI_ > "$env_file"
  docker run $TFCI_DOCKER_ARGS --name tf -w "$TFCI_GIT_DIR" -itd --rm \

  for (int i = 0; i < num_devices; ++i) {
    const string dev_type(TF_DeviceListType(devices, i, status.get()));
    CHECK_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
    const string dev_name(TF_DeviceListName(devices, i, status.get()));
    CHECK_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
    if (dev_type == device_type) {
      *device_name = dev_name;

    how we can reproduce your findings. For memory corruption triggerable in
    TensorFlow models, please demonstrate an exploit against one of Alphabet's
    models in <https://tfhub.dev/>
  - An explanation of who can exploit this vulnerability, and what they gain
    when doing so. Write an attack scenario that demonstrates how your issue