SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf](subq.ai)2 points by mitchwainer 17 days ago | 0 comments