Headless Language Models: Learning without Predicting with Contrastive Weight Tying

Open in new window