Research map / Architectures and efficiency / Quantization
SliderQuant: Accurate Post-Training Quantization for LLMs
Shigeng Wang, Chao Li, Yangyuxuan Kang, Jiawei Fan, Zhonghong Ou, Anbang Yao
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 intra-layer sliding quantization (Intra-S) to fixed-size sliding baseline (ablation) | CBQ-style fixed sliding baseline, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 10.34 | Improved |
| add progressively expanded sliding window (PESW) for shallow layers (ablation) | fixed-size sliding baseline, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 10.30 | Improved |
| add progressively contracted sliding window (PCSW) for deep layers (ablation) | fixed-size sliding baseline, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 9.84 | Improved |
| combine PESW and PCSW (Inter-S) instead of either alone (ablation) | fixed-size sliding baseline, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 9.13 | Improved |
| add Intra-S on top of Inter-S (ablation) | SliderQuant, Llama2-7B · W4A4 | WikiText2 perplexity: 9.13 -> 8.34 | Improved |
| add Intra-S on top of Inter-S (ablation) | SliderQuant, Llama2-7B · W2A16 | WikiText2 perplexity: 10.53 -> 9.59 | Improved |
| increase L_s and L_d from 2 to 6 (ablation) | Inter-S, Llama2-7B · W4A4 | WikiText2 perplexity: 10.23 -> 8.94 | Improved |
| increase fixed window size s from 2 to 4 (no Inter-S) (ablation) | fixed-size sliding baseline, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 11.13 | Improved |
| use 2 stages in Intra-S (gamma 0.5) instead of 1 (ablation) | Intra-S, Llama2-7B · W4A4 | WikiText2 perplexity: 12.73 -> 10.34 | Improved |
| use 4 stages (gamma 0.25) instead of 2 (ablation) | Intra-S, Llama2-7B · W4A4 | WikiText2 perplexity: 10.34 -> 11.32 | Hurt |
| add channel scaling to LoRA in SliderQuant (ablation) | SliderQuant, Llama2-7B · W4A4 | WikiText2 perplexity: 12.95 -> 8.34 | Improved |
| add LoRA to channel scaling in SliderQuant (ablation) | SliderQuant, Llama2-7B · W4A4 | WikiText2 perplexity: 9.18 -> 8.34 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- OmniQuant — OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models (ICLR 2024) · better on 6 of 6 shared results
- FlatQuant — FlatQuant: Flatness Matters for LLM Quantization (ICML 2025) · 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
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.93)
- LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid (ICLR 2025 · similarity 0.92)
- QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models (ICLR 2024 · similarity 0.92)
- CBQ: Cross-Block Quantization for Large Language Models (ICLR 2025 · similarity 0.92)
- Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization (NeurIPS 2025 · similarity 0.91)
- SliM-LLM: Salience-Driven Mixed-Precision Quantization for Large Language Models (ICML 2025 · similarity 0.91)
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.
