size_t proto_len, TF_Status* status) {
  // shape.ParseFromArray takes an int as length, this function takes size_t,
  // make sure there is no information loss.
  if (proto_len > std::numeric_limits<int>::max()) {
    status->status = InvalidArgument(
        "proto_len (", proto_len,
        " bytes) is too large to be parsed by the protocol buffer library");
    return;
  }
  TensorShapeProto shape;

    if (!outputs[i]) continue;

    AbstractTensorHandlePtr numerical_grad;
    {
      AbstractTensorHandle* numerical_grad_raw;
      s = CalcNumericalGrad(ctx, model, inputs,
                            /*input_index=*/i, use_function,
                            &numerical_grad_raw);
      ASSERT_EQ(errors::OK, s.code()) << s.message();
      numerical_grad.reset(numerical_grad_raw);
    }

    TF_Tensor* numerical_tensor;

          BuildImmediateExecutionContext(std::get<1>(GetParam()), &ctx_raw);
      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

          BuildImmediateExecutionContext(std::get<1>(GetParam()), &ctx_raw);
      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

    TF_DeleteTensor(grad_tensor);
    dtheta_approx[i] = grad_data[0];
  }

  // Populate *numerical_grad with the data from dtheta_approx.
  TF_RETURN_IF_ERROR(TestTensorHandleWithDims<float, TF_FLOAT>(
      ctx, dtheta_approx.data(), theta_dims.data(), num_dims, numerical_grad));
  TF_DeleteTensor(theta_tensor);
  return absl::OkStatus();
}

}  // namespace gradients

  Status s;
  AbstractTensorHandlePtr numerical_grad;
  {
    AbstractTensorHandle* numerical_grad_raw;
    s = CalcNumericalGrad(ctx, model, inputs, input_index, use_function,
                          &numerical_grad_raw);
    ASSERT_EQ(errors::OK, s.code()) << s.message();
    numerical_grad.reset(numerical_grad_raw);
  }

  TF_Tensor* numerical_tensor;
  s = GetValue(numerical_grad.get(), &numerical_tensor);

    return tensorflow::ops::Identity(ctx, input, output, name);
  } else if (!DataTypeIsComplex(BaseType(dtype)) &&
             BaseType(dtype) != DT_VARIANT) {
    return errors::InvalidArgument(
        "Expected numeric or variant tensor, got dtype ", dtype);
  }
  return tensorflow::ops::Conj(ctx, input, output, name);
}

class AddGradientFunction : public GradientFunction {
 public:
  Status Compute(AbstractContext* ctx,

          BuildImmediateExecutionContext(std::get<1>(GetParam()), &ctx_raw);
      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }