The results of Part-of-speech tagging is not as accurate as the demo page. What model is used for demo page?

Hello here, I need some help. I used the part-of-speech tagging (PKU) but the result is not as accurate as on the demo page. What is model used in the demo page? How can I make it to achieve the same result?

Case: I have a text “查下实名 mp0000102956 于甜甜” and I want to extract a person name from it.

  • On my code
result = hanlp_mtl_base(text, tasks="pos/pku")


  "tok/fine": [
  "pos/pku": [

Hi, the model weight on our server is proprietary and it is meant to outperform the opensource ones. The only way to access it is our RESTful service.