Why is the calculation of token importance score in generate_training_data.py so inconsistent with that in the paper? #18

Open Mhzzzzz opened 3 years ago