Improving language models by retrieving from trillions of tokens
<h2 class="text-2xl font-bold mb-4">Summary</h2>
This DeepMind paper introduces Retro (Retrieval-augmented Transformer), a language model architecture that improves performance by retrieving informat...
