absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    /* Given upstream grad U and a BiasAdd: A + bias, the gradients are:
     *
     *    dA = U
     *    dbias = reduceSum(U, dims = channel_dim)
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];
    DCHECK(upstream_grad);

  return nullptr;
}

}  // end namespace tracing
}  // end namespace tensorflow

// =============================================================================
// Public C API entry points
//
// These are only the generic entry points for the C API. This file does not
// have any visibility into the graph/eager implementation and is only providing
// C bindings to the abstract classes defined in the

        map sigbuild-r2.12-python3.9 2.12-python3.9
        map sigbuild-r2.12-python3.10 2.12-python3.10
        map sigbuild-r2.12-python3.11 2.12-python3.11
        # TF 2.12 + Clang (containers are the same, but env vars in configs.bzl are different)
        map sigbuild-r2.12-clang 2.12-python3.9
        map sigbuild-r2.12-clang-python3.8 2.12-python3.8
        map sigbuild-r2.12-clang-python3.9 2.12-python3.9

# the dependencies of that target which include a license notice file are then
# included when generating that license. Necessary because the license targets
# in TensorFlow are manually enumerated rather than generated automatically.
do_external_licenses_check(){
  BUILD_TARGET="$1"
  LICENSES_TARGET="$2"

  # grep patterns for targets which are allowed to be missing from the licenses

# 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
  echo "Checking to make sure tensorflow[and-cuda] is installable..."

        ":linux_armhf",
    ],
)

config_setting(
    name = "freebsd",
    values = {"cpu": "freebsd"},
    visibility = ["//visibility:public"],
)

# Features that are default ON are handled differently below.
#
config_setting(
    name = "no_gcp_support",
    define_values = {"no_gcp_support": "true"},
    visibility = ["//visibility:public"],
)

config_setting(

};

// If `persistent_tape` is true, op_tape is not changed and none of the
// backwards functions are deleted.
// If `persistent_tape` is false, op_tape is cleared and backwards functions
// not needed for gradient computation are deleted. Backwards functions that
// are needed, are copied and returned in BackpropInitialState.
template <typename BackwardFunction, typename TapeTensor>

    const char* fn_name = "test_fn";
    core::RefCountPtr<AbstractFunction> scoped_func;
    // Returning null tensors from a tf.function is not supported, so we keep
    // track of indices in the model's outputs are nullptr in this set.
    // The FunctionDef only outputs the non-null tensors. We later pad the
    // function op outputs to have nullptrs at the `null_indices`.
    absl::flat_hash_set<int> null_indices;
    {

# END TF CACHE HELPER OPTIONS
# BEGIN TF TEST SUITE OPTIONS
# These are convenience config options that effectively declare TF's CI test suites. Look
# at the scripts of ci/official/ to see how TF's CI uses them.

# LIBTENSORFLOW TESTS are for building Libtensorflow archives. These are CUDA/CPU-agnostic.

// `src_inputs[i]` will have that input replaced with `dst_inputs[i]`.  `prefix`
// will be prepended to copied node names. `control_deps` are nodes in
// `dst_graph` that the copied `src_graph` nodes will have control dependencies
// on. `return_nodes` are nodes in `src_graph`, and the new corresponding nodes
// in `dst_graph` will be returned. `return_nodes` must be non-null.