TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle(ctx);
  const char remote_device_name[] =
      "/job:localhost/replica:0/task:1/device:CPU:0";
  auto* h0_task1 =
      TFE_TensorHandleCopyToDevice(h0_task0, ctx, remote_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  auto* h1_task1 =
      TFE_TensorHandleCopyToDevice(h1_task0, ctx, remote_device_name, status);

  TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle100x100(ctx);
  const char remote_device_name[] =
      "/job:localhost/replica:0/task:1/device:CPU:0";
  auto* h0_task1 =
      TFE_TensorHandleCopyToDevice(h0_task0, ctx, remote_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  auto* h1_task1 =
      TFE_TensorHandleCopyToDevice(h1_task0, ctx, remote_device_name, status);

  }

  auto* h1_task2 =
      TFE_TensorHandleCopyToDevice(h1_task0, ctx, task2_name, status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

  TFE_TensorHandle* packed_handle = nullptr;
  if (has_packed_input) {
    int num_replicas = 1;
    std::vector<TFE_TensorHandle*> packed_handles = {h1_task2};