Add WIViT instructions to README.md

Per default, the attention values in relation to the first token of the sentence after `[CLS]` are used. You can change this by adjusting line 19 of the script.

### WIViT generation

To generate WIViT, first initialize a model using SBERT_MODEL class and execute the `get_html_word_importance` method on it, with a list of sentence pairs as an argument. A sentence pair should be a tuple of strings.

Example:

```python

ft_model = SBERT_Model("stsb-bert-large", f'{models_path}/stsb-bert-large/', dataset)

ft_model.get_html_word_importance(sentence_pairs)

```

### PAWS evaluation

paws_evaluation.py essentially creates an instance of the SBERT_Model class with all three models and the PAWS data set and calls functions that evaluate those models on it.