Research map / Large language models / Language, safety and interpretability
PASER: Post-Training Data Selection for Efficient Pruned Large Language Model Recovery
Bowei He, Lihao Yin, Hui-Ling Zhen, Xiaokun Zhang, Mingxuan Yuan, Chen Ma
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 semantic-structural clustering (S2RIC) (ablation) | PASER · LLM-Pruner 25% LLaMA2-7B | average reasoning: 59.67 -> 61.10 | Improved |
| add semantic-structural clustering (S2RIC) (ablation) | PASER · SliceGPT 25% LLaMA2-7B | average reasoning: 63.03 -> 64.31 | Improved |
| add capability degradation-aware selection (CDAIS) (ablation) | PASER · LLM-Pruner 25% LLaMA2-7B | average reasoning: 60.26 -> 61.10 | Improved |
| add capability degradation-aware selection (CDAIS) (ablation) | PASER · SliceGPT 25% LLaMA2-7B | average reasoning: 62.68 -> 64.31 | Improved |
| add negative tuning effects mitigation (NTEM) (ablation) | PASER · LLM-Pruner 25% LLaMA2-7B | average reasoning: 59.25 -> 61.10 | Improved |
| add negative tuning effects mitigation (NTEM) (ablation) | PASER · SparseGPT 50% LLaMA2-7B | average reasoning: 61.76 -> 62.78 | Improved |
| recover with full Alpaca data instead of no training | LLM-Pruner 25% · LLaMA2-7B Alpaca | average reasoning: 57.78 -> 35.64 | Hurt |
| select data with PASER instead of Nuggets | LLM-Pruner 25% · LLaMA2-7B Alpaca | average reasoning: 58.69 -> 61.10 | Improved |
| select data with PASER instead of Nuggets | SliceGPT 25% · LLaMA2-7B Alpaca | average reasoning: 59.39 -> 64.31 | Improved |
| select data with PASER instead of Nuggets | SparseGPT 50% · LLaMA2-7B Alpaca | WikiText2 / average: 60.03 -> 62.78 | Improved |
| select data with PASER instead of Nuggets on LLaMA2-70B | LLM-Pruner 50% · LLaMA2-70B | average reasoning: 67.73 -> 69.62 | Improved |
| select data with PASER instead of Nuggets on Baichuan2-7B | LLM-Pruner 25% · Baichuan2-7B | average reasoning: 57.84 -> 59.70 | Improved |
Closest papers by meaning
- An Efficient Pruner for Large Language Model with Theoretical Guarantee (ICML 2025 · similarity 0.91)
- Compact Language Models via Pruning and Knowledge Distillation (NeurIPS 2024 · similarity 0.91)
- Restoring Pruned Large Language Models via Lost Component Compensation (NeurIPS 2025 · similarity 0.91)
- Pruning Large Language Models with Semi-Structural Adaptive Sparse Training (AAAI 2025 · similarity 0.91)
- You Only Prune Once: Designing Calibration-Free Model Compression With Policy Learning (ICLR 2025 · similarity 0.90)
- The LLM Surgeon (ICLR 2024 · 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.
