Breaking Memory Wall for Fast Edge LLM Inference Using Contextual Sparsity

Uses contextual sparsity to get around the memory bandwidth limit in edge LLM inference.