Pre-computed memory or on-the-fly encoding? A hybrid approach to retrieval augmentation makes the most of your compute
This addresses the efficiency problem for users of retrieval-augmented models in knowledge-intensive tasks, offering a cost-effective solution with improved performance, though it is incremental as it builds on existing retrieval methods.
The paper tackled the high computational cost of retrieval-augmented language models by proposing LUMEN, a hybrid approach that pre-computes most retrieval representations and completes encoding on-the-fly, which significantly outperforms pure memory methods on multiple QA tasks while being cheaper than Fusion-in-Decoder, with the advantage increasing with model size.
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.