HEAPr Introduces Atomic Expert Pruning for Efficient LLMs
HEAPr (Hessian-based Efficient Atomic Expert Pruning) trims MoE models by 20%‑25% and cuts FLOPs 20% using atomic expert scores, with two forward passes and one backward pass on a set. Read more: getnews.me/heapr-introduces-atomic-... #heapr #moe #modelpruning