Research map / Architectures and efficiency / Quantization
Principled SVD-based Delta Compression via Quantization Error Minimization
Boya Xiong, Shuo Wang, Weifeng Ge, Guanhua Chen, Yun Chen
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 reconstruction target correction (RTC) (ablation) | PRINMIX · RTC ablation (AVG) | AVG accuracy: 60.2 -> 61.5 | Improved |
| add RTC (ablation) | PRINMIX · AIME2024 (R1-Distill-Qwen-14B) | accuracy: 27.5 -> 31.1 | Improved |
| add RTC (ablation) | PRINMIX · Math500 (R1-Distill-Qwen-14B) | accuracy: 78.2 -> 80.2 | Improved |
| add RTC (ablation) | PRINMIX · LLAVA-1.5 GQA and SQA | accuracy: 62.8 vs 62.7 on GQA; 72.2 vs 72.1 on SQA | No clear effect |
| vary fmax from 2 to 6 (ablation) | PRINMIX · fmax sweep, R1-Distill-Qwen-14B | AVG accuracy: 57.0, 55.0, 55.7, 56.4, 56.4 vs 50.5 for Delta-CoMe | No clear effect |
| replace Delta-CoMe's mixed precision with ILP quantization-error minimization | PRINMIX · 13-14B models AVG | AVG accuracy: 63.3 -> 64.7 | Improved |
| replace Delta-CoMe's mixed precision with ILP quantization-error minimization | PRINMIX · 7B models AVG | AVG accuracy: 71.9 -> 74.0 | Improved |
| replace Delta-CoMe with PRINMIX | PRINMIX · AIME2024 (R1-Distill-Qwen-7B) | accuracy: 30.0 -> 36.7 | Improved |
| replace Delta-CoMe with PRINMIX | PRINMIX · MBPP/HumanEval 7B coder | accuracy: 82.7 vs 83.1 on MBPP; 85.0 vs 85.6 on Humaneval | No clear effect |
| replace Delta-CoMe with PRINMIX | PRINMIX · MoE/enc-dec/Qwen3 mix AVG | AVG accuracy: 42.9 -> 44.5 | Improved |
| compress delta with PRINMIX instead of LoRA at alpha=1/16 | PRINMIX · DeepSeek-LLM-7B code+math AVG | AVG accuracy: 35.5 -> 40.8 | Improved |
| apply delta compression (PRINMIX) instead of PRINMIX-LoRA at alpha=1/64 | PRINMIX · DeepSeek-LLM-7B code+math AVG | AVG accuracy: 35.2 -> 38.4 | Improved |
Closest papers by meaning
- Delta-CoMe: Training-Free Delta-Compression with Mixed-Precision for Large Language Models (NeurIPS 2024 · similarity 0.92)
- SVD-LLM: Truncation-aware Singular Value Decomposition for Large Language Model Compression (ICLR 2025 · similarity 0.88)
- Dobi-SVD: Differentiable SVD for LLM Compression and Some New Perspectives (ICLR 2025 · similarity 0.88)
- PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression (NeurIPS 2024 · similarity 0.88)
- LowRankArena: A Standardized Evaluation Platform for SVD-Based LLM Compression (NeurIPS 2026 · similarity 0.88)
- QERA: an Analytical Framework for Quantization Error Reconstruction (ICLR 2025 · similarity 0.87)
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.
