New sota on conll03 data?

kyzhouhzau / BERT-NER

Use Google's BERT for named entity recognition （CoNLL-2003 as the dataset）.

MIT License

1.24k stars 335 forks source link

New sota on conll03 data? #2

Open qiuwei opened 5 years ago

qiuwei commented 5 years ago

hi, so the F1 of your system is 0.9326? Is it comparable with the numbers reported in the literature?

kyzhouhzau commented 5 years ago

@qiuwei In their report, the F score on the development set is 0.964. The result of 0.932 is lower than them, but I also found that the results of their own MRPC codes lower than the values reported in their paper , so i think it must has some tricks on parameter adjustment .

qiuwei commented 5 years ago

Thanks for the information. I thought the number was on the test set. Did you use the large model or the base model?

kyzhouhzau commented 5 years ago

@qiuwei

the result was on the dev, i used base model ,when i use large i got OOM.