reimplement xlm-r about mpt

[lifu-tu]

Could you have a check again? I corrected typos in the script. I just confirmed that the exact match score is ~82-83% after 1 epoch training. You can also delete some process data cache in case your last data processing is broken.

[nuochenpku]

Thanks for your advice, I find it's caused by the lr. In fact, lr should be the 3e-5.
But now I have another concern about the experimental results of udpos. Could you provide the whole data for udpos, I cannot get the test set of udpos from Xtreme.

[lifu-tu]

The udpos data is added. Please remember to cite the original dataset paper if you use it. Thanks!

@Article{nivre2018universal,
title={Universal Dependencies 2.2},
author={Nivre, Joakim and Abrams, Mitchell and Agi{'c}, {\v{Z}}eljko and Ahrenberg, Lars and Antonsen, Lene and Aranzabe, Maria Jesus and Arutie, Gashaw and Asahara, Masayuki and Ateyah, Luma and Attia, Mohammed and others},
year={2018}
}

[nuochenpku]

Of course, thanks for your response!