[BUG] Metrics throwing new error on master since 0.4.0

[jbragg]

The following error occurs when passing in references of type List[List[str]] to metrics like bleu.
Wasn't happening on 0.4.0 but happening now on master.

  File "/usr/local/lib/python3.7/site-packages/nlp/metric.py", line 226, in compute
    self.add_batch(predictions=predictions, references=references)
  File "/usr/local/lib/python3.7/site-packages/nlp/metric.py", line 242, in add_batch
    batch = self.info.features.encode_batch(batch)
  File "/usr/local/lib/python3.7/site-packages/nlp/features.py", line 527, in encode_batch
    encoded_batch[key] = [encode_nested_example(self[key], cast_to_python_objects(obj)) for obj in column]
  File "/usr/local/lib/python3.7/site-packages/nlp/features.py", line 527, in <listcomp>
    encoded_batch[key] = [encode_nested_example(self[key], cast_to_python_objects(obj)) for obj in column]
  File "/usr/local/lib/python3.7/site-packages/nlp/features.py", line 456, in encode_nested_example
    raise ValueError("Got a string but expected a list instead: '{}'".format(obj))

[jbragg]

Update - maybe this is only failing on bleu because I was not tokenizing inputs to the metric

[jbragg]

Closing - seems to be just forgetting to tokenize. And found the helpful discussion in huggingface/evaluate#105