Could you please provide more details about the word-based bert? about uer-py HOT 2 CLOSED

Thank you for your attention
We trained word-based BERT upon wikizh and retained 80 thousands words according to frequency. The pre-trained model and vocabulary are publicly available (see Chinese model zoo section).
All low-frequency words are mapped o [UNK]. You can change the model according to your vocabulary by using dynamic_vocab_adapter.py (see scripts section).
You have multiple choices to find similar words, including the use of cloze_test.py , topn_words_indep.py , and topn_words_dep.py . We hope the instructions in scripts section are really helpful to you.

from uer-py.

Thank you for your reply! I'll read the scripts more carefully.

from uer-py.