This is a small demo that explores the use of seq2seq
to highlight interesting parts of texts, where interesting is
defined as unexpected to gpt2. You may launch it with
python perplexity_main.py
Here is a picture:
The dependencies are: transformers, pytorch, numpy and gradio.