Changing from gpt to gpt2

I get an error when I change from gpt to gpt2 in the interactive.py file.
However, when I change from gpt to gpt2 in the train.py file, everything is fine.

The error I get:
File "/transfer-learning-conv-ai/interact.py", line 130, in run
logger.info("Selected personality: %s", tokenizer.decode(chain(*personality)))

File "/pytorch_transformers/tokenization_utils.py", line 767, in decode
sub_texts.append(self.convert_tokens_to_string(current_sub_text))

File "/pytorch_transformers/tokenization_gpt2.py", line 198, in convert_tokens_to_string
text = ''.join(tokens)

TypeError: sequence item 0: expected str instance, NoneType found

I also tried to switch from pytorch-transformer to the new transformers library to no avail.

Compare the GPT and GPT-2 paths in interactive.py and train.py, then trace the failure from interact.py line 130 into pytorch_transformers/tokenization_utils.py and tokenization_gpt2.py. Reproduce the interactive GPT-2 run and confirm it completes without the NoneType decoding error.

Written by the indexing model from the issue text.