Model Pruning

This is a Claude Code template for compressing large language models through pruning techniques like Wanda and SparseGPT. You'd reach for it when deploying models to edge devices or trying to cut inference costs without full retraining. The promised 40-60% size reduction with under 1% accuracy loss is impressive if it holds up in practice, and the 2-4x speedup matters for production serving. It's based on solid research (ICLR 2024 papers), though with mixed security audit results you'll want to review the code before running it on anything sensitive. The 27K stars suggest the broader template collection has traction, but only 353 installs on this specific skill means you're in early adopter territory.