K-bert cannot achieve performance as paper mention

I use k-bert run_ner.py but cannot get the result as high as the paper say, and I dont know how to change dataset to finetune

Start with run_ner.py and compare the reported K-BERT result with the paper and the attached result. Determine which dataset and fine-tuning configuration were used, then document a reproducible setup and whether it reaches the claimed performance.

Written by the indexing model from the issue text.