48 Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models · 17 authors 4