Neph0s/LMKE

UMLS reulsts

Closed this issue · 2 comments

NoviceCrom commented
UMLS reulsts
NoviceCrom commented

Could you share your best hyper-parameters on UMLS like learning_rate etc?

Neph0s commented

With
BERT-base, we set the learning rate of PLMs and other components as 10−5
and 5×10−4
respectively. For triple classification, we set the batch size as 16 and sample 1 negative triple
for each positive triple.