Research map / Architectures and efficiency / Pruning and efficient tuning
ZipLM: Inference-Aware Structured Pruning of Language Models
Eldar Kurtic, Elias Frantar, Dan Alistarh
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 |
|---|---|---|---|
| prune with ZipLM instead of DistilGPT2 at similar speedup (throughput) | GPT2 · WikiText-103 | perplexity: 43.0 -> 35.4 | Improved |
| prune for latency with ZipLM instead of DistilGPT2 | GPT2 · WikiText-103 | perplexity: 43.0 -> 41.2 | Improved |
| ZipLM one-shot vs Kwon et al., 2.0x speedup | BERT-base · SQuAD | F1: 76.5 -> 84.1 | Improved |
| ZipLM one-shot vs Kwon et al., 2.0x speedup | BERT-base · QQP | accuracy: 83.9 -> 84.8 | Improved |
| ZipLM one-shot vs Kwon et al., 2.0x speedup | BERT-base · MNLI | accuracy: 78.1 -> 78.2 | No clear effect |
| prune for speedup instead of for sparsity (ablation) | ZipLM · BERT-base | accuracy: up to 10 points | Improved |
| increase calibration samples from 4 to 32 (ablation) | ZipLM one-shot · SQuAD1.1, 2.0x | F1: 48.4 -> 82.6 | Improved |
| increase calibration samples from 32 to 4096 (ablation) | ZipLM one-shot · SQuAD1.1, 1.5x | F1: 86.8 -> 87.6 | Improved |
| add layer-wise token distillation to ZipLM (ablation) | ZipLM · SST-2 | accuracy: 89.2 -> 91.7 | Improved |
| add layer-wise token distillation to ZipLM (ablation) | ZipLM · QNLI | accuracy: 86.5 -> 88.6 | Improved |
| add layer-wise token distillation to ZipLM (ablation) | ZipLM · SQuAD | F1: 85.7 -> 85.7 | No clear effect |
| use ZipLM instead of CoFi with distillation | BERT-base · SQuAD | F1: 82.6 -> 85.7 | Improved |
Closest papers by meaning
- SparseLLM: Towards Global Pruning of Pre-trained Language Models (NeurIPS 2024 · similarity 0.91)
- SlimLLM: Accurate Structured Pruning for Large Language Models (ICML 2025 · similarity 0.91)
- SeedLM: Compressing LLM Weights into Seeds of Pseudo-Random Generators (ICLR 2025 · similarity 0.90)
- Compact Language Models via Pruning and Knowledge Distillation (NeurIPS 2024 · similarity 0.90)
- Fluctuation-Based Adaptive Structured Pruning for Large Language Models (AAAI 2024 · similarity 0.90)
- DLP: Dynamic Layerwise Pruning in Large Language Models (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.
