Research map / Architectures and efficiency / Quantization
ReSpinQuant: Efficient Layer-Wise LLM Quantization via Subspace Residual Rotation Approximation
Suyoung Kim, Sunghyun Wee, Hyeonjin Kim, Kyomin Hwang, Hyunho Lee, Nojun Kwak
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 rank-8 residual subspace rotation (vs identity, rank 0) (ablation) | ReSpinQuant · WikiText-2 W3A3 LLaMA-3 8B | perplexity: 20.03 -> 14.20 | Improved |
| increase rank from 8 to 32 (ablation) | ReSpinQuant · WikiText-2 W3A3 LLaMA-3 8B | perplexity: 14.20 -> 13.09 | Improved |
| increase rank from 32 to 128 (ablation) | ReSpinQuant · WikiText-2 W3A3 LLaMA-3 8B | perplexity: 13.09 -> 12.80 | Improved |
| use full-rank (4096) residual instead of rank 32 (ablation) | ReSpinQuant · WikiText-2 W3A3 LLaMA-3 8B | perplexity: 13.09 -> 12.52 | Improved |
| increase rank from 64 to 128 (ablation) | ReSpinQuant · WikiText-2 W3A3 LLaMA-3 8B | zero-shot accuracy: 51.29 -> 50.80 | Hurt |
| add layer-wise rotations with residual subspace correction to SpinQuant | SpinQuant · WikiText-2 W4A4 LLaMA-3 8B | perplexity: 7.50 -> 7.24 | Improved |
| add layer-wise rotations to SpinQuant | SpinQuant · WikiText-2 W3A3 LLaMA-3.2 1B | perplexity: 69.70 -> 49.90 | Improved |
| use layer-wise rotation instead of global rotation | SpinQuant · LLaMA-3 8B calibration time | minutes: 17 -> 42 | Hurt |
| use residual rotation instead of global rotation | SpinQuant · LLaMA-3 8B W4A4 batch 16 decoding | TTIT (ms): 160.95 -> 163.81 | Hurt |
| add online low-rank rotation r=32 | SpinQuant · LLaMA-3 8B MACs | additional MACs: 15.4M -> 32.3M | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- SpinQuant — SpinQuant: LLM Quantization with Learned Rotations (ICLR 2025) · better on 5 of 7 shared results
- OSTQuant — OSTQuant: Refining Large Language Model Quantization with Orthogonal and Scaling Transformations for Better Distribution Fitting (ICLR 2025) · better on 5 of 6 shared results
- FlatQuant — FlatQuant: Flatness Matters for LLM Quantization (ICML 2025) · better on 4 of 4 shared results
- QuaRot — QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs (NeurIPS 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SpinQuant: LLM Quantization with Learned Rotations (ICLR 2025 · similarity 0.90)
- ResQ: Mixed-Precision Quantization of Large Language Models with Low-Rank Residuals (ICML 2025 · similarity 0.90)
- ConQuR: Corner Aligned Activation Quantization via Optimized Rotations for LLMs (NeurIPS 2026 · similarity 0.90)
- LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid (ICLR 2025 · similarity 0.89)
- DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs (NeurIPS 2024 · similarity 0.89)
- Rotated Runtime Smooth: Training-Free Activation Smoother for accurate INT4 inference (ICLR 2025 · similarity 0.89)
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.
