Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
The Pantagraph

Bloomington-Normal to implement parking bans for snow removal

Bloomington and Normal have implemented parking bans to go into effect just as snow is expected to begin falling across Central Illinois.