Efficient Attention via Pre-Scoring: Prioritizing Informative Keys in Transformers

Open in new window