SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf](subq.ai)
2 pointsby mitchwainer2 hours ago1 comment