AI Paper Atlas

Research map / Architectures and efficiency / Quantization

ProjQ: Project-and-Quantize for Adapter-Aware LLM Compression

Wenya Yu, Chao Zhang, Li Wang, Samson Lasaulce, Merouane Abdelkader DEBBAH

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 0 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
use ProjQ instead of LoftQLLaMA2-7B 2-bit · C4perplexity: 28.77 -> 21.50Improved
use ProjQ instead of GPTQ+SVD-LLMLLaMA2-7B 2-bit · Common Sense avgaccuracy (%): 45.33 -> 47.82Improved
use ProjQ instead of LoftQ before fine-tuningLLaMA2-7B 2-bit · WikiText-2 fine-tunedperplexity: 9.14 -> 8.17Improved
use ProjQ instead of LoftQLLaMA2-7B 2-bit · Common Sense fine-tunedaccuracy (%): 54.78 -> 57.59Improved
use ProjQ instead of LoftQLLaMA2-7B 3-bit · GSM8K fine-tunedaccuracy (%): 37.68 -> 35.18Hurt
use ProjQ instead of LoftQQwen2.5-14B-Ins 2-bit · GSM8K fine-tunedaccuracy (%): 51.25 -> 51.11No clear effect
use ProjQ instead of QLoRAQwen2.5-7B-Ins 4-bit · Common Sense fine-tunedaccuracy (%): 71.59 -> 66.78Hurt
add proposed adapter initialization over standard LoRA init (ablation)ProjQ LLaMA2-7B · WikiText-2 fine-tunedperplexity: 9.17 -> 8.17Improved
add proposed adapter initialization over standard LoRA init (ablation)ProjQ Qwen2.5-7B-Ins · Common Sense avgaccuracy (%): 53.66 -> 59.88Improved
increase alternating iterations from 1 to 3 (ablation)ProjQ LLaMA2-7B · C4perplexity: 25.62 -> 25.32No clear effect
increase alternating iterations from 1 to 2 (ablation)ProjQ LLaMA2-7B · Common Sense avgaccuracy (%): 46.91 -> 40.91Hurt
set designed rank r_d equal to adapter rank 64 instead of 16 (ablation)ProjQ LLaMA2-7B · WikiText (zero-shot)perplexity: 29.07 -> 22.42Improved

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