Tag: PrismML

PrismML bets reasoning models can shrink onto phones

The Caltech spinout squeezed an Alibaba model to 5.9GB while keeping 98 percent of its benchmark scores.