Timezone: »

Pre-computed memory or on-the-fly encoding? A hybrid approach to retrieval augmentation makes the most of your compute
Michiel de Jong · Yury Zemlyanskiy · Nicholas FitzGerald · Joshua Ainslie · Sumit Sanghai · Fei Sha · William Cohen

Wed Jul 26 02:00 PM -- 03:30 PM (PDT) @ Exhibit Hall 1 #306

Retrieval-augmented language models such as Fusion-in-Decoder are powerful, setting the state of the art on a variety of knowledge-intensive tasks. However, they are also expensive, due to the need to encode a large number of retrieved passages. Some work avoids this cost by pre-encoding a text corpus into a memory and retrieving dense representations directly. However, pre-encoding memory incurs a severe quality penalty as the memory representations are not conditioned on the current input. We propose LUMEN, a hybrid between these two extremes, pre-computing the majority of the retrieval representation and completing the encoding on the fly using a live encoder that is conditioned on the question and fine-tuned for the task. We show that LUMEN significantly outperforms pure memory on multiple question-answering tasks while being much cheaper than FiD, and outperforms both for any given compute budget. Moreover, the advantage of LUMEN over FiD increases with model size.

Author Information

Michiel de Jong (University of Southern California)
Yury Zemlyanskiy (Google)
Nicholas FitzGerald (Google DeepMind)
Joshua Ainslie (Google)
Sumit Sanghai (Research, Google)
Fei Sha (Google Research)
William Cohen (Google AI)

More from the Same Authors