Fix minor error-handling bug (tensorflow#8468)

tensorflow/contrib/seq2seq/python/ops/helper.py

@@ -443,7 +443,7 @@ def sample(self, time, outputs, state, name=None):
     # Outputs are logits, use argmax to get the most probable id
     if not isinstance(outputs, ops.Tensor):
       raise TypeError("Expected outputs to be a single Tensor, got: %s" %
-                      outputs)
+                      type(outputs))
     sample_ids = math_ops.cast(
         math_ops.argmax(outputs, axis=-1), dtypes.int32)
     return sample_ids