# Sourcing this enables local disk cache
if [[ $(uname -s) == "Darwin" ]]; then
  echo "Please note that using disk cache on macOS is not recommended because the"
  echo "cache can end up being pretty big and make the build process inefficient."
fi

#   export TF_ANY_TARGETS="quoted list of targets, like on the command line"
#   export TF_ANY_MODE=test
set -euxo pipefail
cd "$(dirname "$0")/../../"  # tensorflow/
export TFCI="$(echo $TFCI | sed 's/,nightly_upload/,public_cache,disk_cache/')"
git bisect start "$TF_BISECT_BAD" "$TF_BISECT_GOOD"

# Sourcing this enables Bazel remote cache (read and write)
# Note that "_push" cache configs write to GCS buckets and require
# authentication. If you are not a Googler, source "public_cache" to enable the
# public read-only cache.
# The cache configs are different for MacOS and Linux
if [[ $(uname -s) == "Darwin" ]]; then
  TFCI_BAZEL_COMMON_ARGS="$TFCI_BAZEL_COMMON_ARGS --config tf_public_macos_cache_push"
else

#      Ex. macos_arm64      -- arm64 MacOS platform
# 3. Add modifiers. Some modifiers for local execution are:
#      Ex. disk_cache -- Use a local cache
#      Ex. public_cache -- Use TF's public cache (read-only)
#      Ex. public_cache_push -- Use TF's public cache (read and write, Googlers only)
#      Ex. rbe        -- Use RBE for faster builds (Googlers only; see below)

}

Status ImmediateExecutionTensorHandle::SummarizeValue(
    std::string& summary) const {
  Status status;
  AbstractTensorPtr resolved(
      // TODO(allenl): Resolve should be const, and the caches that get updated
      // marked mutable.
      const_cast<ImmediateExecutionTensorHandle*>(this)->Resolve(&status));
  if (!status.ok()) {
    return status;
  }
  summary = resolved->SummarizeValue();

# Sourcing this enables Bazel remote cache (public, read-only)
# The cache configs are different for MacOS and Linux
if [[ $(uname -s) == "Darwin" ]]; then
  TFCI_BAZEL_COMMON_ARGS="$TFCI_BAZEL_COMMON_ARGS --config tf_public_macos_cache"
else
  TFCI_BAZEL_COMMON_ARGS="$TFCI_BAZEL_COMMON_ARGS --config tf_public_cache"

// is initialized (either locally or remotely). The context_id can change during
// the process lifetime although this should cause the worker to be
// reinitialized (e.g. cleared caches) as well.
TF_CAPI_EXPORT extern uint64_t TFE_GetContextId(TFE_Context* ctx);

// -----------------------------------------------------------------------------
// Cancellation APIs.

  docker pull "$TFCI_DOCKER_IMAGE" || sleep 15
  docker pull "$TFCI_DOCKER_IMAGE" || sleep 15
  docker pull "$TFCI_DOCKER_IMAGE"
fi 

if [[ "$TFCI_DOCKER_REBUILD_ENABLE" == 1 ]]; then
  DOCKER_BUILDKIT=1 docker build --cache-from "$TFCI_DOCKER_IMAGE" -t "$TFCI_DOCKER_IMAGE" $TFCI_DOCKER_REBUILD_ARGS
  if [[ "$TFCI_DOCKER_REBUILD_UPLOAD_ENABLE" == 1 ]]; then
    docker push "$TFCI_DOCKER_IMAGE"
  fi
fi

        "//tensorflow/core/distributed_runtime/rpc:grpc_channel",
        "//tensorflow/core/distributed_runtime/rpc:grpc_server_lib",
        "//tensorflow/core/distributed_runtime/rpc:grpc_worker_cache",
        "//tensorflow/core/distributed_runtime/rpc:grpc_worker_service",
        "//tensorflow/core/distributed_runtime/rpc:rpc_rendezvous_mgr",
        "//tensorflow/core/distributed_runtime/rpc/eager:grpc_eager_client",

    same implementation as their `tf.losses` equivalent.
*   For Keras model, the individual call of `Model.evaluate` uses no cached data
    for evaluation, while `Model.fit` uses cached data when `validation_data`
    arg is provided for better performance.
*   Adds a `save_traces` argument to `model.save`/ `tf.keras.models.save_model`