* support adding keywords to model tokenizer * add keyword_preprocess.py usage doc * init new token with good weights --------- Co-authored-by: Charles Ju <charlesyju@gmail.com>
charlesyju