Research map / Architectures and efficiency / Quantization
No outlier channels but with outlier blocks
Shanwen Mao, Hao Zhang, Jiasheng Li, Haoyu Qiao, Chenxin Cai, Tingting Wu, Jie Liu
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 OCP (ablation) | AQLM (3-bit) · WikiText2, Llama3-8B | perplexity: 7.05 -> 6.93 | Improved |
| add OCP (ablation) | VPTQ (3-bit) · WikiText2, Llama3-8B | perplexity: 7.01 -> 6.86 | Improved |
| add OCP (ablation) | GPTVQ (3-bit) · WikiText2, Llama3-8B | perplexity: 6.03 -> 6.12 | Hurt |
| add OCP (ablation) | GPTVQ (2-bit) · WikiText2, Llama3-8B | perplexity: 10.34 -> 9.14 | Improved |
| add OCP (ablation) | VPTQ (2-bit) · WikiText2, Llama3-8B | perplexity: 9.19 -> 8.92 | Improved |
| add OCP (ablation) | AQLM (2-bit) · MMLU average, Llama3-8B | accuracy: 44.67 -> 45.63 | Improved |
| add OCP (ablation) | GPTVQ (3-bit) · MMLU average, Llama3-8B | accuracy: 50.71 -> 48.81 | Hurt |
| use linear-level compensation instead of random (ablation) | NuBitQ · layer 31 of LLaMA3-8B | relative PPL reduction: 1.00× -> 5.27× | Improved |
| use model-level compensation instead of random (ablation) | NuBitQ · layer 31 of LLaMA3-8B | relative PPL reduction: 1.00× -> 2.12× | Improved |
| use NuBitQ with OCP instead of NuBitQ alone at 2-bit scale | NuBitQ · WikiText2, Llama3-8B | perplexity: 5.79 -> 5.66 | Mixed |
| increase vector length d from 8 to 64 (ablation) | NuBitQ · Ada Lovelace computation time | time (ms): 63.7 -> 98.8 | Hurt |
| shrink vector length to d=4 versus FP16 (ablation) | NuBitQ · Ada Lovelace computation time | time (ms): 134.3 -> 63.6 | Improved |
Closest papers by meaning
- Rethinking Channel Dimensions to Isolate Outliers for Low-bit Weight Quantization of Large Language Models (ICLR 2024 · similarity 0.91)
- DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs (NeurIPS 2024 · similarity 0.90)
- ASER: Activation Smoothing and Error Reconstruction for Large Language Model Quantization (AAAI 2025 · similarity 0.88)
- FlatQuant: Flatness Matters for LLM Quantization (ICML 2025 · similarity 0.87)
- Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models (AAAI 2024 · similarity 0.87)
- Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment (NeurIPS 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.
