// Calculate Grad
    std::string name = "relu_grad";
    TF_RETURN_IF_ERROR(ReluGrad(ctx, upstream_grad, activations,
                                &grad_inputs[0], name.c_str()));
    return absl::OkStatus();
  }
  ~ReluGradientFunction() override {
    for (auto output : forward_outputs_) {
      if (output) {
        output->Unref();
      }
    }
  }

 private:

  TFE_TensorHandle* input_eager =
      TestTensorHandleWithDims<T, datatype>(eager_ctx, data, dims, num_dims);
  *tensor =
      unwrap(TF_CreateAbstractTensorFromEagerTensor(input_eager, status.get()));
  return absl::OkStatus();
}

// Return a scalar tensor handle with given value.
template <class T, TF_DataType datatype>
Status TestScalarTensorHandle(AbstractContext* ctx, const 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);
    return absl::OkStatus();
  }

    DCHECK(grad_outputs[0]);
    grad_inputs[0] = grad_outputs[0];
    grad_inputs[1] = grad_outputs[0];

    grad_inputs[0]->Ref();
    grad_inputs[1]->Ref();
    return absl::OkStatus();
  }
  ~AddGradientFunction() override {}
};

class ExpGradientFunction : public GradientFunction {
 public:
  explicit ExpGradientFunction(AbstractTensorHandle* exp) : exp_(exp) {
    exp->Ref();
  }

                   tensor_interface->ByteSize()))) {
      return InvalidArgument(
          "Malformed TF_RESOURCE tensor: unable to parse resource handle");
    }
    return absl::OkStatus();
  }
  return absl::OkStatus();
}

static bool TF_Run_Inputs(TF_Tensor* const* c_inputs,
                          std::vector<std::pair<string, Tensor>>* input_pairs,
                          TF_Status* status) {

    TF_RETURN_IF_ERROR(input->Shape(&shape));
    TF_RETURN_IF_ERROR(dyn_cast<tracing::TracingContext>(ctx)->AddParameter(
        input->DataType(), shape, &handle));
    params->emplace_back(handle);
  }
  return absl::OkStatus();
}

// Runs `model` maybe wrapped in a function.
Status RunModel(Model model, AbstractContext* ctx,
                absl::Span<AbstractTensorHandle* const> inputs,

  // Although, `ops::SparseSoftmaxCrossEntropyWithLogits` returns 2 tensors, the
  // second tensor isn't needed for computing gradient so we could safely drop
  // it.
  outputs[0] = loss;
  backprop->Unref();
  return absl::OkStatus();
}

Status BiasAddModel(AbstractContext* ctx,
                    absl::Span<AbstractTensorHandle* const> inputs,
                    absl::Span<AbstractTensorHandle*> outputs) {

    CHECK_EQ(grad_outputs.size(), 1);
    CHECK_EQ(grad_inputs.size(), 1);
    grad_inputs[0] = grad_outputs[0];
    if (grad_inputs[0]) {
      grad_inputs[0]->Ref();
    }
    return absl::OkStatus();
  }
};

// Computes:
//
// @tf.custom_gradient
// def f(input):
//   def grad(grads):
//     return grads[0]
//   return tf.exp(input), grad
// outputs = [f(inputs[0])]

  return IsRefType(dt)
             ? InvalidArgument("Output ", idx, " of node '", node->name(),
                               "' has a reference type ", DataTypeString(dt))
             : absl::OkStatus();
}

// Converts `ninputs` and `inputs` into `inputs_tensors` and `input_nodes` and
// does various checks while doing so. `input_nodes` will contain the same

      func_record_(new FunctionRecord(std::move(fdef), {}, true)) {}
GraphFunction::~GraphFunction() {}
Status GraphFunction::GetFunctionDef(const FunctionDef **fdef) {
  *fdef = &(func_record_->fdef());
  return absl::OkStatus();
}
}  // namespace graph
}  // namespace tracing