on Ampere based GPUs. TensorFloat-32, or TF32 for short, is a math mode for
    NVIDIA Ampere based GPUs and is enabled by default.

*   A major refactoring of the internals of the Keras Functional API has been
    completed, that should improve the reliability, stability, and performance
    of constructing Functional models.

*   Keras mixed precision API

# These you may need to change for your own GCP project.
common:rbe_linux_cpu --remote_instance_name=projects/tensorflow-testing/instances/default_instance

# TODO(kanglan): Remove it after toolchain update is complete.
build:rbe_linux_cpu_old --config=rbe_linux
build:rbe_linux_cpu_old --host_crosstool_top="@ubuntu20.04-gcc9_manylinux2014-cuda11.2-cudnn8.1-tensorrt7.2_config_cuda//crosstool:toolchain"