AI Paper Atlas

Research map / Architectures and efficiency / Quantization

PrefixQuant: Eliminating Outliers by Prefixed Tokens for Large Language Models Quantization

Mengzhao Chen, Yi Liu, Jiahao Wang, Yi Bin, Wenqi Shao, Ping Luo

TPAMI 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 35 citations (OpenAlex)

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 changedWhereMeasuredEffect
add prefixed outlier tokens in KV cache (dynamic) (ablation)Llama-3-8B W4A4KV4 with rotation and grid search · WikiText2perplexity: 11.70 -> 7.53Improved
add prefixed outlier tokens in KV cache (static) (ablation)Llama-3-8B W4A4KV4 static quantization · WikiText2perplexity: 141.02 -> 7.93Improved
add block-wise fine-tuning (static) (ablation)PrefixQuant-O2 Llama-3-8B W4A4KV4 · WikiText2perplexity: 7.93 -> 7.41Improved
add block-wise fine-tuning (dynamic) (ablation)PrefixQuant-O1 Llama-3-8B W4A4KV4 · WikiText2perplexity: 7.53 -> 7.23Improved
add Hadamard rotation to RTN (ablation)Llama-3-8B W4A4KV4 dynamic · WikiText2perplexity: 1282.34 -> 24.98Improved
add grid search initialization (ablation)Llama-3-8B W4A4KV4 rotated · WikiText2perplexity: 24.98 -> 11.70Improved
use adaptively computed number of prefixed tokens instead of none (ablation)PrefixQuant Llama-2-7B W4A4KV4 O2 · WikiText2perplexity: 17.63 -> 6.01Improved
use top-N high-frequency tokens instead of random prefix tokens (ablation)PrefixQuant w/o FT Llama-2-7B · WikiText2perplexity: 66.51 -> 6.22Improved
use top-N tokens instead of only the highest-frequency token (ablation)PrefixQuant w/o FT Llama-2-7B · WikiText2perplexity: 12.07 -> 6.22Improved
use Pile instead of C4 as fine-tuning dataset (ablation)PrefixQuant-O2 Llama-3-8B W4A4KV4 · WikiText2perplexity: 7.60 -> 7.42Improved
fine-tune 20 epochs instead of none (ablation)PrefixQuant-O2 Llama-3-8B W4A4KV4 · WikiText2perplexity: 7.93 -> 7.42Improved
increase fine-tuning tokens from 256x1024 to 1024x1024 (ablation)PrefixQuant-O2 Llama-3-8B W4A4KV4 · WikiText2perplexity: 7.46 -> 7.41No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most 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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy