Cropping outperforms dropout as an augmentation strategy for training self-supervised text embeddings

Open in new window