AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

CGSVD: Cascaded Granular Singular Value Decomposition for Large Language Model Compression

Yuli Chen, Shuhao Zhang, Jiale Han, Fanshen Meng, Haishen Jiang, Bo Cheng, Qiang Tong, Xiulei Liu

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data

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 changedWhereMeasuredEffect
use CGSVD cascaded non-uniform allocation instead of uniform SVD-LLMSVD-LLM · LLaMA3.1-8B, 30% compressionWikiText2 perplexity: 32.92 -> 17.46Improved
use CGSVD instead of SVD-LLM at 60% compressionSVD-LLM · LLaMA2-7B WikiText2perplexity: 89.95 -> 49.92Improved
add layer importance via angular distance (inter-layer allocation) (ablation)SVD-LLM uniform allocation · LLaMA2-7B, WikiText2, 60%perplexity: 89.95 -> 65.85Improved
add module compressibility via spectral entropy (intra-layer allocation) (ablation)SVD-LLM uniform allocation · LLaMA2-7B, WikiText2, 60%perplexity: 89.95 -> 73.83Improved
combine layer importance and module compressibility (ablation)SVD-LLM uniform allocation · LLaMA2-7B, WikiText2, 60%perplexity: 89.95 -> 50.48Improved
add Iterative Residual Filling (ablation)CGSVD (LI + MC) · LLaMA2-7B, WikiText2, 60%perplexity: 50.48 -> 49.92Improved
add Iterative Residual Filling (ablation)CGSVD (LI + MC) · LLaMA2-7B, WikiText2, 30%perplexity: 9.56 -> 9.37Improved
add LoRA fine-tuning to CGSVDCGSVD · LLaMA2-7B, WikiText2, 60%perplexity: 49.92 -> 16.72Improved
increase calibration sequence length (ablation)CGSVD · WikiText2perplexity: 20.28 -> 8.65Improved
increase calibration batch size (ablation)CGSVD · WikiText2perplexity: 11.31 -> 9.29Improved
add 8-bit GPTQ quantization on top of CGSVDCGSVD · LLaMA2-7B WikiText2, 30%perplexity: 9.37 -> 9.37No clear effect
add 4-bit GPTQ quantization on top of CGSVDCGSVD · LLaMA2-7B WikiText2, 30%perplexity: 9.37 -> 14.31Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy