SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf] subq.ai 2 points by mitchwainer 25 days ago · 1 comment Reader PiP Save No comments yet.