Hacker News new | ask | show | jobs
SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf] (subq.ai)
2 points by mitchwainer 46 days ago