Pre-computed memory or on-the-fly encoding? A hybrid approach to retrieval augmentation makes the most of your compute

Open in new window