std::vector<int64_t> source_tensor_ids = MakeTensorIDList(sources);
  tensorflow::gtl::FlatSet<int64_t> sources_set(source_tensor_ids.begin(),
                                                source_tensor_ids.end());
  std::unordered_map<int64_t, TapeTensor> sources_that_are_targets;
  for (int i = 0; i < target_tensor_ids.size(); ++i) {
    int64_t target_id = target_tensor_ids[i];

  TFE_TensorHandle* (*pack)(TFE_Context* context, TFE_TensorHandle** handles,
                            int num_handles, TF_Status* s,
                            void* device_info) = nullptr;

  // Pins the op to `device` based on inputs to `op`. Returns true
  // signifying to pin to the current custom device. Returns false
  // to pin to the physical device.
  //

* `tf.tpu.experimental.embedding.TPUEmbeddingV2`
    * Add `compute_sparse_core_stats` for sparse core users to profile the 
      data with this API to get the `max_ids` and `max_unique_ids`. These
      numbers will be needed to configure the sparse core embedding mid level 
      api.
    * Remove the `preprocess_features` method since that's no longer needed.

## Thanks to our Contributors

  }
  if (result.size() != source_tensor_ids.size()) {
    return errors::Internal("Expected result Span to be of size ",
                            source_tensor_ids.size(), " found ", result.size(),
                            " in call to Tape::ComputeGradient.");
  }
  std::unordered_set<int64_t> used_gradient_ids(source_tensor_ids.size());
  for (int i = 0; i < source_tensor_ids.size(); i++) {