Research map / Architectures and efficiency / Quantization
SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| apply SmoothQuant-O3 W8A8 quantization | OPT-175B · zero-shot average | accuracy: 66.9% -> 66.8% | No clear effect |
| use SmoothQuant-O3 instead of W8A8 naive quantization | OPT-175B · zero-shot average | accuracy: 35.5% -> 66.8% | Helped |
| use SmoothQuant-O3 instead of ZeroQuant | OPT-175B · zero-shot average | accuracy: 35.8% -> 66.8% | Helped |
| use SmoothQuant-O3 instead of Outlier Suppression | OPT-175B · zero-shot average | accuracy: 36.0% -> 66.8% | Helped |
| use SmoothQuant-O3 instead of LLM.int8() | OPT-175B · zero-shot average | accuracy: 66.7% -> 66.8% | No clear effect |
| apply SmoothQuant-O3 W8A8 quantization | OPT-13B · sequence length 256, batch 4 | latency: 152.6 -> 112.1 | Helped |
| apply W8A8 quantization with SmoothQuant | MT-NLG 530B · zero-shot average | accuracy: 73.1% -> 73.1% | No clear effect |
| use O3 instead of O1 quantization (ablation) | SmoothQuant on OPT-30B · sequence length 256, batch 4 | latency: 246.7 -> 227.6 | Helped |
| use O3 instead of O2 quantization (ablation) | SmoothQuant on OPT-30B · sequence length 256, batch 4 | latency: 240.2 -> 227.6 | Helped |
| use O3 quantization rather than O1 (ablation) | SmoothQuant on OPT-175B · LAMBADA | accuracy: 74.7% -> 74.6% | No clear effect |
| use O3 quantization rather than O1 (ablation) | SmoothQuant on OPT-175B · WikiText | perplexity: 11.11 -> 11.17 | Hurt |
| choose migration strength alpha in the 0.4-0.6 sweet spot rather than below 0.4 or above 0.6 (ablation) | SmoothQuant on OPT-175B · LAMBADA | accuracy | Helped |
Closest papers by meaning
- LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid (ICLR 2025 · similarity 0.93)
- ASER: Activation Smoothing and Error Reconstruction for Large Language Model Quantization (AAAI 2025 · similarity 0.92)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.92)
- QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models (ICLR 2024 · similarity 0.92)
- OSTQuant: Refining Large Language Model Quantization with Orthogonal and Scaling Transformations for Better Distribution Fitting (ICLR 2025 · similarity 0.92)
- ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models (AAAI 2025 · similarity 0.92)
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.