Wisp+

New pruning methods enhance LLM efficiency by preserving output differences

Researchers have introduced a new family of pruning methods called "difference-informed pruning" designed to improve the efficiency of large language models. These methods focus on preserving the differences between mod…