I want to run simcse (contrastive learning with dropout as data augmentation for NLP). How do I change the dropout mask in forward pass, so that I get two embeddings: one with an older dropout mask, one with a newer dropout mask?
I want to run simcse (contrastive learning with dropout as data augmentation for NLP). How do I change the dropout mask in forward pass, so that I get two embeddings: one with an older dropout mask, one with a newer dropout mask?