SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf]

 (subq.ai)

2 points | by mitchwainer 5 hours ago ago

1 comments