6 ms·SubQ: Sub-quadratic LLM built for 12M-token reasoning4 points by anujbans 4mo agomartypitt 4mo agoImpressive benchmarks!