ResearchBenchmarksLLM Pruning Evaluation on Math Dimension GSM8K MinervaLLM Pruning Evaluation on Math Dimension GSM8K MinervaFollow68.1Relative Retention (Math)SparseGPT48.298453.439258.5863.7208Jun 2, 2026Evaluation resultsMethodMethodLinksRelative Retention (Math)SparseGPTCalibration=MetaMath,...Calibration=MetaMath, Model Scale=LLaMA-3.1-70B, Sparsity=60% unstructured2026.0668.1SparseGPTCalibration=GSM8K, Mod...Calibration=GSM8K, Model Scale=LLaMA-3.1-8B, Sparsity=60% unstructured2026.0649.06