Griffin's Compromise: Combining Gated Linear Recurrence with Local Attention
This article separates the evidence reported by the paper from the open questions of adoption, explaining how Hawk and Griffin combine recurrent-model efficiency with the Transformer's representational power.