Hyper parameters for ChildTuning

[jiangllan]

Thanks a lot for all the details you provide in Appendix B for reproducibility! However, I still encounter some difficulties in reproducing the experiment.
I noticed that you apply grid search. Could you please provide the specific , and learning rate for each task?

[RunxinXu]

Cola pf=pd=0.2
Rte pf=pd=0.3
Mrpc pf=0.4 pd=0.1
Stab pf=0.2 pd=0.2

It would be better to grid search the hyper-parameters.

[jiangllan]

thx!!