Research map / Architectures and efficiency / Pruning and efficient tuning
Prune&Comp: Free Lunch for Layer-Pruned LLMs via Iterative Pruning with Magnitude Compensation
Xinrui Chen, Hongxing Zhang, Fanyi Zeng, Yongxian Wei, Yizhi Wang, Xitong Ling, Guanghao Li, Chun Yuan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add iterative pruning to one-shot pruning (ablation) | CosSim(BI), LLaMA-3-8B 7/32 · PPL average | perplexity: 58.43 -> 38.37 | Improved |
| add magnitude compensation to one-shot pruning (ablation) | CosSim(BI), LLaMA-3-8B 7/32 · PPL average | perplexity: 58.43 -> 37.39 | Improved |
| add magnitude compensation to iterative pruning (ablation) | CosSim(BI), LLaMA-3-8B 7/32 · PPL average | perplexity: 38.37 -> 27.16 | Improved |
| add Prune&Comp to Taylor+ | LLaMA-3-8B 5/32 · PPL average | perplexity: 512.78 -> 16.34 | Improved |
| add Prune&Comp to Taylor+ | LLaMA-3-8B 7/32 · PPL average | perplexity: 2840.38 -> 25.21 | Improved |
| add Prune&Comp to CosSim(CL) | LLaMA-3-8B 7/32 · PPL average | perplexity: 2839.30 -> 230.73 | Improved |
| add Prune&Comp to Mag+ | LLaMA-2-7B 7/32 · PPL average | perplexity: 89.61 -> 28.20 | Improved |
| add Prune&Comp to Taylor+ | LLaMA-3-8B 5/32 · QA relative performance | relative performance: 65.85 -> 90.57 | Improved |
| add Prune&Comp to CosSim(BI) | LLaMA-3-8B 5/32 · QA relative performance | relative performance: 89.18 -> 93.19 | Improved |
| add Prune&Comp to CosSim(CL) | Qwen3-8B 5/36 · QA relative performance | relative performance: 84.42 -> 89.29 | Improved |
| add Prune&Comp to PPL metric | Qwen3-8B 5/36 · QA relative performance | relative performance: 91.98 -> 89.22 | Hurt |
| add Prune&Comp to Taylor+ | LLaMA-3-8B 5/32 · MMLU | weighted accuracy: 38.64 -> 48.42 | Improved |
Closest papers by meaning
- Layer as Puzzle Pieces: Compressing Large Language Models through Layer Concatenation (NeurIPS 2025 · similarity 0.93)
- Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization (NeurIPS 2025 · similarity 0.92)
- SlimGPT: Layer-wise Structured Pruning for Large Language Models (NeurIPS 2024 · similarity 0.91)
- The Unreasonable Ineffectiveness of the Deeper Layers (ICLR 2025 · similarity 0.90)
- AlphaPruning: Using Heavy-Tailed Self Regularization Theory for Improved Layer-wise Pruning of Large Language Models (NeurIPS 2024 · similarity 0.90)
- An Efficient Pruner for Large Language Model with Theoretical Guarantee (ICML 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
