Research map / Large language models / Language, safety and interpretability
Universal Properties of Activation Sparsity in Modern Large Language Models
Filip Szatkowski, Patryk Będkowski, Alessio Devoto, Jan Dubiński, Pasquale Minervini, Mikołaj Piórczyński, Simone Scardapane, Bartosz Wójcik
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 |
|---|---|---|---|
| scale Gemma3 from 1B to 27B | Gemma3 pretrained, intermediate activations · Intermediate sparsification | critical sparsity (%): 50.22 -> 74.12 | Improved |
| scale Gemma3 from 1B to 27B | Gemma3 pretrained, all inputs · All-Inputs sparsification | critical sparsity (%): 35.15 -> 66.26 | Improved |
| scale LLaMA from 3.2-1B to 3.1-8B | LLaMA pretrained · Intermediate sparsification | critical sparsity (%): 44.44 -> 51.89 | Improved |
| scale Qwen2.5 from 0.5B to 32B | Qwen2.5 pretrained · All-Inputs sparsification | critical sparsity (%): 36.66 -> 58.93 | Improved |
| scale Qwen2.5 from 3B to 7B | Qwen2.5 pretrained · Intermediate sparsification | critical sparsity (%): 71.16 -> 60.98 | Hurt |
| instruction-tune Gemma3-27B | Gemma3-27B · Intermediate sparsification | critical sparsity (%): 74.12 -> 84.05 | Improved |
| instruction-tune Gemma3-12B | Gemma3-12B · All-Inputs sparsification | critical sparsity (%): 52.54 -> 61.45 | Improved |
| instruction-tune Gemma3-1B | Gemma3-1B · Intermediate sparsification | critical sparsity (%): 50.22 -> 49.98 | No clear effect |
| instruction-tune Qwen2.5-0.5B | Qwen2.5-0.5B · Intermediate sparsification | critical sparsity (%): 46.54 -> 43.92 | Hurt |
| sparsify intermediate activations instead of all FFN inputs | LLaDA-8B · Average over tasks | critical sparsity (%): 56.79 -> 68.13 | Improved |
| sparsify intermediate activations instead of gate in Gemma3-27B | Gemma3-27B pretrained · critical sparsity | critical sparsity (%): 53.03 -> 74.12 | Improved |
| use top-k or max-p instead of top-p (ablation) | sparsification rule · sparsity-accuracy tradeoff | critical sparsity (%) | No clear effect |
Closest papers by meaning
- Sparsing Law: Towards Large Language Models with Greater Activation Sparsity (ICML 2025 · similarity 0.95)
- R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference (ICLR 2025 · similarity 0.92)
- Training-Free Activation Sparsity in Large Language Models (ICLR 2025 · similarity 0.91)
- Adaptive Layer Sparsity for Large Language Models via Activation Correlation Assessment (NeurIPS 2024 · similarity 0.91)
- ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models (ICLR 2024 · similarity 0.91)
- Learn To be Efficient: Build Structured Sparsity in Large Language Models (NeurIPS 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.
