wall-clock speedup

SparseDecoding: Pruning Guided by Decoding for Accurate and Efficient LLM Inference

SparseDecoding is a decoding-aware LLM pruning framework for faster inference. It reports 1.48x wall-clock spe…