Research map / Architectures and efficiency / Pruning and efficient tuning
Efficient Plug-and-Play Weight Refinement for Sparse Large Models
Jingcheng Xie, Yinda Chen, Xiaoyu Liu, Yinglong Li, Haoyuan Shi, Zhiwei Xiong
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 EIWR refinement | SparseGPT · WikiText2 LLaMA2-7B 70% | perplexity: 23.5870 -> 20.2348 | Improved |
| add EIWR refinement | Wanda · WikiText2 LLaMA2-7B 70% | perplexity: 73.0853 -> 22.9877 | Improved |
| add EIWR refinement | Wanda · WikiText2 Qwen2.5-3B 70% | perplexity: 152.2175 -> 37.0856 | Improved |
| add PCG refinement instead of EIWR | SparseGPT · WikiText2 LLaMA2-7B 70% | perplexity: 23.5870 -> 21.9632 | Improved |
| add DSnoT refinement | SparseGPT · WikiText2 LLaMA2-7B 70% | perplexity: 23.5870 -> 67.5221 | Hurt |
| add DSnoT refinement | Wanda · WikiText2 LLaMA2-7B 70% | perplexity: 73.0853 -> 72.9401 | No clear effect |
| add EIWR at 90% sparsity | SparseGPT · WikiText2 LLaMA2-7B 90% | perplexity: 1.32e3 -> 673.23 | Improved |
| add EIWR at 80% sparsity | SparseGPT · WikiText2 LLaMA3-8B 80% | perplexity: 176.42 -> 137.60 | Improved |
| add DSnoT on top of EIWR (ablation) | EIWR · WikiText2 LLaMA2-7B 70% | perplexity: 20.23 -> 19.73 | Improved |
| add AlphaPruning on top of EIWR (ablation) | EIWR · WikiText2 LLaMA2-7B 70% | perplexity: 20.23 -> 17.81 | Improved |
| use progressive instead of no momentum (ablation) | EIWR · WikiText2 LLaMA2-7B 70% | perplexity: 20.61 -> 16.47 | Improved |
| use progressive instead of fixed momentum (ablation) | EIWR · WikiText2 LLaMA2-7B 70% | perplexity: 18.95 -> 16.47 | Improved |
Closest papers by meaning
- Targeted Low-rank Refinement: Enhancing Sparse Language Models with Precision (ICML 2025 · similarity 0.90)
- The Journey Matters: Average Parameter Count over Pre-training Unifies Sparse and Dense Scaling Laws (ICLR 2025 · similarity 0.89)
- A Simple and Effective Pruning Approach for Large Language Models (ICLR 2024 · similarity 0.89)
- Pruning Large Language Models with Semi-Structural Adaptive Sparse Training (AAAI 2025 · similarity 0.89)
- Dynamic Sparse No Training: Training-Free Fine-tuning for Sparse LLMs (ICLR 2024 · similarity 0.89)
- APT: Adaptive Pruning and Tuning Pretrained Language Models for Efficient Training and Inference (ICML 2024 · similarity 0.88)
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.
