```

TensorFlow kernels and TensorFlow's dependencies are still not built with
debugging information with `--config=dbg`, as issues occur on Linux if
there is too much debug info (see [this GitHub
issue](https://github.com/tensorflow/tensorflow/issues/48919) for context). If
you want to debug a kernel, you can compile specific files with `-g` using the

# it enables a derived --config setting.  If RBE is not available (i.e. there
# is no --config setting), bazel would fail and quit. This script does a quick
# check This script checks for such errors early
if ! grep "rbe_$TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX" .bazelrc; then
  cat <<EOF
ERROR: RBE was enabled via the 'rbe' env in the 'TFCI' variable.
       TFCI: $TFCI

export TF_VER_PATCH=$(awk '/#define TF_PATCH_VERSION/ {print $3}' tensorflow/core/public/version.h)

# Note: in awk, "print $N" prints the Nth "field", where fields are strings separated
# by whitespace. The flag "-F<x>" changes the behavior such that fields are now strings
# separated by the character <x>. Therefore, -F\' (escaped single quote) means that field
# $2 in <Tensor'flow'> is <flow>. This is useful for reading string literals like below.

# specifically. Use your best judgment to keep the scripts in this directory
# lean and easy to follow. When in doubt, remember that for CI scripts, "keep it
# simple" is MUCH more important than "don't repeat yourself."

# -e: abort script if one command fails
# -u: error if undefined variable used
# -x: log all commands
# -o pipefail: entire command fails if pipe fails. watch out for yes | ...

#   your builds. This usually saves a lot of time, especially when
#   re-running tests. However, note that:
#
#    - New environments like new CUDA versions, changes to manylinux,
#      compilers, etc. can cause undefined behavior such as build failures
#      or tests passing incorrectly.
#    - Automatic LLVM updates are known to extend build time even with
#      the cache; this is unavoidable.
export TFCI=py311,linux_x86,public_cache,disk_cache

    int64_t init_timeout_in_ms, TF_Status* status,
    bool clear_existing_contexts);

// Set server def with retries and timeout. This is helpful for fault-tolerant
// initial connection in high-preemption environments, such as
// ParameterServerStrategy training.
// This API is for experimental usage and may be subject to change.
TF_CAPI_EXPORT extern void TFE_ContextSetServerDefWithTimeoutAndRetries(

        on Ampere based GPUs.TensorFloat-32, or TF32 for short, is a math mode
        for NVIDIA Ampere based GPUs which causes certain float32 ops, such as
        matrix multiplications and convolutions, to run much faster on Ampere
        GPUs but with reduced precision. This reduced precision has not been
        found to effect convergence quality of deep learning models in practice.

        return;
      }
      // We have a convoluted scheme here: Using the C++ graph construction API
      // to add potentially many nodes to the graph without running the checks
      // (such as uniqueness of the names of nodes) we run with other functions
      // that add a node to the graph (like TF_FinishOperation).
      if (!g->name_map.insert(std::make_pair(n->name(), n)).second) {

# New rules should be added above this target.
# -------------------------------------------

# TensorFlow uses several libraries that may also be used by applications
# linking against the C and C++ APIs (such as libjpeg).  When we create
# the shared library, only export the core TF API functions to avoid
# causing library conflicts (e.g., those reported in github issue 1924).

    std::unordered_map<int64_t, OpTapeEntry<BackwardFunction, TapeTensor>>;

// Operations the tape needs to perform on tensors to do backpropagation. Named
// "vspace" because a subset of these are related to a vector space, such as
// adding gradients, getting zeroes, etc. Currently cannot be implemented
// without using tensorflow python code, hence left unspecified here.
//
// Gradient is the type returned by gradient functions. In Python TF it's either