Don't Judge a Language Model by Its Last Layer: Contrastive Learning with Layer-Wise Attention Pooling

Dongsuk Oh, Yejin Kim, Hodong Lee, H. Howie Huang, Heuiseok Lim