TF_EXPECT_OK(ReadCache(&cache, "", block_size, 1, &out));
  TF_EXPECT_OK(ReadCache(&cache, "", block_size, 1, &out));
  calls.push_back(2 * block_size);
  // Cache miss followed by cache hit.  Causes eviction of LRU element.
  TF_EXPECT_OK(ReadCache(&cache, "", 2 * block_size, 1, &out));
  TF_EXPECT_OK(ReadCache(&cache, "", 2 * block_size, 1, &out));
  // LRU element was at offset 0.  Cache miss.
  calls.push_back(0);

            with `mask_zero=True` to automatically infer a correct padding mask.
        *   Added a `use_causal_mask` call time argument to the layer. Passing
            `use_causal_mask=True` will compute a causal attention mask, and
            optionally combine it with any `attention_mask` passed in directly
            when calling the layer.

      const char* attr_name, absl::Span<const AbstractOperation*> values) = 0;

 private:
  const AbstractOperationKind kind_;
};

// TODO(b/193656009): Defining these in a cc file causes linker errors with
// fastbuild.
inline Status AbstractOperation::SetAttrShape(const char* attr_name,
                                              const PartialTensorShape shape) {

#   variables in between script executions. To forcibly remove the
#   container and start fresh, run "docker rm -f tf". Removing the container
#   destroys some temporary bazel data and causes longer builds.
#
#   You will need the NVIDIA Container Toolkit for GPU testing:
#   https://github.com/NVIDIA/nvidia-container-toolkit
#
#   Note: if you interrupt a bazel command on docker (ctrl-c), you

// If assume_valid_feeds is true, it can help infer shapes in the fanout of fed
// nodes. This may cause incorrectness in graph analyses, but is useful for
// simulation or scheduling.
// If aggressive_shape_inference is true, nodes are executed on the host to
// identify output values when possible and does other aggressive strategies.
// This may cause incorrectness in graph analyses, but is useful for simulation
// or scheduling.

// make sure all nodes are finished.
TF_CAPI_EXPORT extern void TFE_DeleteExecutor(TFE_Executor*);

// Returns true if the executor is in async mode.
TF_CAPI_EXPORT extern bool TFE_ExecutorIsAsync(TFE_Executor*);

// Causes the calling thread to block till all ops dispatched in this executor
// have been executed. Note that "execution" here refers to kernel execution /
// scheduling of copies, etc. Similar to sync execution, it doesn't guarantee

    **Note:** You can add `--test_sharding_strategy=disabled` to the `flags` to
    disable the sharding so that all the test outputs are in one file. However,
    it may slow down the tests for not running in parallel and may cause the
    test to timeout but it could be useful when you need to execute a single
    test or more in general your filtered/selected tests have a very low
    execution time and the sharding

   ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
   LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
   ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

implement, and therefore security issues deriving from their absence are not
considered a vulnerability in TensorFlow.

### Resource allocation

A denial of service caused by one model could bring down the entire server, but
we don't consider this as a vulnerability, given that models can exhaust
resources in many different ways and solutions exist to prevent this from

/// different compiler options breaks compatibility. For more details about how
/// this is used, please consult next sections. Here we just wrap these tables
/// in lint warnings so that changes here cause changes to the versioning data
/// as well. Here is a short summary of what changes are allowed:
///   * adding a new method at the end of a table is allowed at any time;