Recurrent Memory Transformer retains information across up to 2 million tokens (words). Applying Transformers to long texts does not necessarily require large amounts of memory. By employing a ...