tokenizer = bert_base_chinese for config and model you can always use bert_base_chinese