SubQ: a sub-quadratic LLM with 12M-token context 3 months agoLooks like long context isn’t a problem anymore 0ThreadHN