3 ms·SubQ: Sub-quadratic LLM built for 12M-token reasoning4 points by anujbans 5mo agomartypitt 5mo agoImpressive benchmarks!