Large AI models often contain parts that contribute little to their answers. Pruning identifies and removes these redundant connections, producing a leaner model that runs faster and uses less memory while keeping most of its ability.
Along with other efficiency methods, pruning helps deploy capable models on limited hardware and cuts the cost and energy of running them. Prune too aggressively and quality suffers, so it is done carefully and the result is tested against the original. It is one of several complementary techniques that together make modern models practical to run at scale.