Replies: 1 comment 2 replies
-
Hi @lxmm1999, kindly try using the following setting and post your experience. ##### WebQSP
--relation_dim 200
--que_embedding_model SentenceTransformer
--do_batch_norm 0
--freeze 1
--batch_size 16
--validate_every 10
--lr 0.00002
--entdrop 0.0
--reldrop 0.0
--scoredrop 0.0
--decay 1.0
--model ComplEx
--patience 20
--ls 0.0
--l3_reg 0.001
--nb_epochs 200 |
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi,
You mentioned here that SBert performed better than Roberta.I tried a few times but couldn't get better results.
Are the hyperparameters the same as Roberta when using SBert? I would appreciate it if you could provide hyperparameters that you used to get a better result.
Beta Was this translation helpful? Give feedback.
All reactions