ASSERT_EQ(errors::OK, s.code()) << s.message();
    s = dyn_cast<tracing::TracingContext>(ctx.get())->AddParameter(
        DT_FLOAT, shape, &x_raw);
    ASSERT_EQ(errors::OK, s.code()) << s.message();
    x.reset(x_raw);
  }

  PartialTensorShape shape;
  Status s = x->Shape(&shape);
  ASSERT_EQ(errors::OK, s.code()) << s.message();
  ASSERT_FALSE(shape.unknown_rank());

TensorFlow process effectively executes arbitrary code.

The risk of loading untrusted checkpoints depends on the code or graph that you
are working with. When loading untrusted checkpoints, the values of the traced
variables from your model are also going to be untrusted. That means that if
your code interacts with the filesystem, network, etc. and uses checkpointed

fi
# VERY basic check to ensure the [and-cuda] package variant is installable.
# Checks TFCI_BAZEL_COMMON_ARGS for "gpu" or "cuda", implying that the test is
# relevant. All of the GPU test machines have CUDA installed via other means,
# so I am not sure how to verify that the dependencies themselves are valid for
# the moment.
if [[ "$TFCI_BAZEL_COMMON_ARGS" =~ gpu|cuda ]]; then

load(
    "//tensorflow:tensorflow.bzl",
    "tf_cc_test",
    "tf_copts",
    "tf_cuda_cc_test",
    "tf_cuda_library",
)
load("//tensorflow:tensorflow.default.bzl", "cc_header_only_library", "filegroup")
load(
    "//tensorflow/core/platform:build_config_root.bzl",
    "tf_cuda_tests_tags",
)
load("//tensorflow/core/platform:rules_cc.bzl", "cc_library")

package(

# limitations under the License.
# ==============================================================================
source "${BASH_SOURCE%/*}/utilities/setup.sh"

# Record GPU count and CUDA version status
if [[ "$TFCI_NVIDIA_SMI_ENABLE" == 1 ]]; then
  tfrun nvidia-smi
fi

# Update the version numbers for Nightly only
if [[ "$TFCI_NIGHTLY_UPDATE_VERSION_ENABLE" == 1 ]]; then

# limitations under the License.
# ==============================================================================
source "${BASH_SOURCE%/*}/utilities/setup.sh"

# Record GPU count and CUDA version status
if [[ "$TFCI_NVIDIA_SMI_ENABLE" == 1 ]]; then
  tfrun nvidia-smi
fi

# Update the version numbers for Nightly only
if [[ "$TFCI_NIGHTLY_UPDATE_VERSION_ENABLE" == 1 ]]; then

    TF_SetStatusFromGCSStatus(gcs_status, status);
    auto status_code = TF_GetCode(status);
    if (status_code != TF_OK && status_code != TF_NOT_FOUND) return;
    // If this file does not exist on server, we will need to sync it.
    bool sync_need = (status_code == TF_NOT_FOUND);
    file->plugin_file = new tf_writable_file::GCSFile(

  for (size_t i = 0; i < tasks.length; ++i) {
    const auto& result = (*results)[i];
    TF_Status s;
    TF_SetStatus(&s, static_cast<TF_Code>(result.error_code()),
                 std::string(result.error_message()).data());
    if (TF_GetCode(&s) != TF_Code::TF_OK) {
      tensorflow::CoordinationServiceError error;
      *error.mutable_source_task() = result.error_payload().source_task();

// removal of the registration part from `Env` and adding it here instead: we
// will need tables to hold for each scheme the function tables that implement
// the needed functionality instead of the current `FileSystemRegistry` code in
// `core/platform/env.cc`.
class ModularFileSystem final : public FileSystem {
 public:
  ModularFileSystem(
      std::unique_ptr<TF_Filesystem> filesystem,
      std::unique_ptr<const TF_FilesystemOps> filesystem_ops,

  ImmediateExecutionOperation* CreateOperation() override = 0;

  // Returns whether the runtime is backed by TFRT or the legacy TF Eager
  // Runtime. This is necessary to decouple runtime-dependent
  // code that is layered on top of the runtime.
  virtual bool UsesTFRT() = 0;

  // List attributes of available devices
  virtual void ListDevices(std::vector<DeviceAttributes>* devices) = 0;