const char* value,
                                                    TF_Status* status);

// Get configuration key and value using coordination service.
// The config key must be set before getting its value. Getting value of
// non-existing config keys will result in errors.
// If `timeout_in_ms=0`, this call will block until the key-value is set or the
// worker shuts down.

                   ForwardOperation* forward_op_) {
  forward_op_->attrs.Set(attr_name, value);
  return op_->SetAttrBool(attr_name, value);
}
Status SetAttrType(AbstractOperation* op_, const char* attr_name,
                   DataType value, ForwardOperation* forward_op_) {
  forward_op_->attrs.Set(attr_name, value);
  return op_->SetAttrType(attr_name, value);
}

  auto s = tensorflow::unwrap(op)->SetAttrFloat(attr_name, value);
  if (!s.ok()) {
    LOG(WARNING) << "Unable to set attribute: " << attr_name;
  }
}

void TFE_OpSetAttrBool(TFE_Op* op, const char* attr_name, unsigned char value) {
  auto s = tensorflow::unwrap(op)->SetAttrBool(attr_name,
                                               (value == 0) ? false : true);
  if (!s.ok()) {

  } else {
    // Block until the key-value is set or the worker shuts down.
    timeout = absl::InfiniteDuration();
  }
  auto status_or_value = coord_agent->GetKeyValue(key, timeout);
  status->status = status_or_value.status();
  if (!status_or_value.ok()) return;

  const std::string& value_string = status_or_value.value();
  void* data = tensorflow::port::Malloc(value_string.length());

  memcpy(TF_TensorData(t.get()), &value, TF_TensorByteSize(t.get()));

  std::unique_ptr<TFE_TensorHandle, decltype(&TFE_DeleteTensorHandle)>
      value_handle(TFE_NewTensorHandle(t.get(), status),
                   TFE_DeleteTensorHandle);
  if (TF_GetCode(status) != TF_OK) return nullptr;

  TFE_OpAddInput(op, value_handle.get(), status);

  }
  Status SetAttrInt(const char* attr_name, int64_t value) override {
    op_->node_builder.Attr(attr_name, static_cast<int64_t>(value));
    return absl::OkStatus();
  }
  Status SetAttrFloat(const char* attr_name, float value) override {
    op_->node_builder.Attr(attr_name, value);
    return absl::OkStatus();
  }
  Status SetAttrBool(const char* attr_name, bool value) override {
    op_->node_builder.Attr(attr_name, value);

  DLTensor* dl_tensor = &dlmt->dl_tensor;
  absl::optional<std::string> device_name =
      DeviceNameFromDlContext(dl_tensor->device, status);
  if (!device_name.has_value()) {
    status->status =
        tensorflow::errors::InvalidArgument("Unsupported Device Type");
    return nullptr;
  }
  TF_DataType dtype;
  Status s = TfDataTypeFormDlDataType(dl_tensor->dtype, &dtype);

                                    TF_Status* status) {
  status->status = value.status();
  if (!value.ok()) {
    return nullptr;
  }
  // Caller is responsible to call `TF_DeleteBuffer` to release the buffer.
  TF_Buffer* result = TF_NewBuffer();
  const std::string& value_str = *value;
  void* data = malloc(value_str.length());

  // unique_ptrs to ensure this happens.
  std::function<void(BackwardFunction*)> backward_function_deleter;
};

// Map from tensor_id to internally-defined operation-id of the operation which
// produced this tensor. A value of -1 means that the tensor was directly
// watched and not the result of any operation in the tape.
using TensorTape = std::unordered_map<int64_t, int64_t>;

// Map from operation-id to tape entry.

      "      attr {"
      "        key: 'dtype'"
      "        value {"
      "          type: DT_FLOAT"
      "        }"
      "      }"
      "    }"
      "    node_def {"
      "      name: 'add1'"
      "      op: 'Add'"
      "      input: 'read0:value:0'"
      "      input: 'read1:value:0'"
      "      attr {"
      "        key: 'T'"
      "        value {"
      "          type: DT_FLOAT"
      "        }"