"@local_config_cuda//cuda:curand + "\
    "@local_config_cuda//cuda:cusolver + "\
    "@local_config_tensorrt//:tensorrt)" --keep_going > $BATS_TEST_TMPDIR/out

  cat <<EOF
There was a path found connecting //tensorflow/tools/pip_package:build_pip_package
to a banned CUDA dependency. Here's the output from bazel query:
EOF
  cat $BATS_TEST_TMPDIR/out
  [[ ! -s $BATS_TEST_TMPDIR/out ]]
}

    *   Add SeparableConv1D layer.
    *   Add convolutional Flipout layers.
    *   When both inputs of `tf.matmul` are bfloat16, it returns bfloat16,
        instead of float32.
    *   Added `tf.contrib.image.connected_components`.
    *   Add `tf.contrib.framework.CriticalSection` that allows atomic variable
        access.
    *   Output variance over trees predictions for classifications tasks.

    "@local_config_cuda//cuda:curand + "\
    "@local_config_cuda//cuda:cusolver + "\
    "@local_config_tensorrt//:tensorrt)" --keep_going > $BATS_TEST_TMPDIR/out

  cat <<EOF
There was a path found connecting //tensorflow/tools/pip_package:wheel
to a banned CUDA dependency. Here's the output from bazel query:
EOF
  cat $BATS_TEST_TMPDIR/out
  [[ ! -s $BATS_TEST_TMPDIR/out ]]
}

        // operations which supposedly both created the same Tensor. It comes up
        // in recompute_grad, where the gradients have the same value. However,
        // only the original gradient is connected to everything else, so we
        // should still use that.
        vspace_.DeleteGradient(forward_grads[i]);
      } else {
        accumulated_gradients_[output_tensors[i].GetID()] = forward_grads[i];
      }