fix TransformerQA predictor test

tests/rc/predictors/transformer_qa_test.py

@@ -20,12 +20,10 @@ def test_predict_single_instance(self):
  "What is love?", "Baby don't hurt me, don't hurt me, no more."
  )
  span_start, span_end = prediction["best_span"]
- assert 0 <= span_start <= span_end
- assert (
- "best_span_str" in prediction
- and isinstance(prediction["best_span_str"], str)
- and len(prediction["best_span_str"]) > 0
- )
+ assert -1 <= span_start <= span_end
+ assert "best_span_str" in prediction and isinstance(prediction["best_span_str"], str)
+ if span_start > -1:
+ assert len(prediction["best_span_str"]) > 0
 
  def test_predict_long_instance(self):
  # We use a short context and a long context, so that the long context has to be broken into multiple