// workloads, which is required since the Python implementation depends on the
// above mutation methods. This allows us to prevent modifications to nodes in
// the graph after the session has been made aware of them.
TF_CAPI_EXPORT extern void TF_ExtendSession(TF_Session* session,
                                            TF_Status* status);

// Returns the serialized CppShapeInferenceResult::HandleData proto for

# limitations under the License.
# ==============================================================================
#
# Usage: rename_and_verify_wheels.sh
# This script is aware of TFCI_ variables, so it doesn't need any arguments.
# Puts new wheel through auditwheel to rename and verify it, deletes the old
# one, checks the filesize, and then ensures the new wheel is installable.
set -euxo pipefail

happening (e.g., rate limits, ACLs, monitors to restart broken servers).

### Model sharing

If the multitenant design allows sharing models, make sure that tenants and
users are aware of the security risks detailed here and that they are going to
be practically running code provided by other users. Currently there are no good
ways to detect malicious models/graphs/checkpoints, so the recommended way to

        training loop.
    *   **Performance regression on `ParameterServerStrategy`.** This could be
        significant if you have many PS servers. We are aware of this issue and
        working on fixes, for now we suggest using the legacy optimizers when
        using `ParameterServerStrategy`.
    *   **Timeout or performance loss.** We don't anticipate this to happen, but