NVIDIA's DeltaNet-2 improves efficient long-context language model memory

Original: This is a very interesting paper and architecture.

Source: x.com ↗