Reproduce “Prune Gradually, Adapt Sparsely: Efficient Pruning of Large Language Models at Extreme Sparsity” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jean Lucien Randrianantenaina, Hung Nguyen, Nasib Ullah, Jinbin Zhang, Nilesh Gupta, Inderjit S. Dhillon, Rohit Babbar
ETAPS progressively prunes language models, adapting a small set of surviving weights between cuts and using gradients to guide later selections. This staged approach is evaluated on LLaMA-2-7B at 90% sparsity, where it achieves lower WikiText-2 and C4 perplexity than ELSA, with fewer aggregate GPU-hours and less persistent state. The comparison does not show faster wall-clock pruning or inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.