Prompt Compression and Contrastive Conditioning for Controllability and Toxicity Reduction in Language Models

Open in new window