AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs

Yoonjun Cho, Dongjae Jeon, Soeun Kim, Moongyu Jeon, Albert No

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 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
preserve top-k singular subspace before quantizing (SRR) on QERA-exactQERA-exact, 3-bit MXINT r=64 · WikiText2 Gemma-2 2Bperplexity: 19.36 -> 18.30Improved
add SRR to LQERLQER, 3-bit MXINT r=32 · WikiText2 LLaMA-2 7Bperplexity: 14.51 -> 11.22Improved
add SRR to QERA-exactQERA-exact, 3-bit MXINT r=64 · WikiText2 LLaMA-3.1 70Bperplexity: 6.55 -> 6.46Improved
add SRR to QERA-exactQERA-exact, 3-bit MXINT r=64 · 5-task zero-shot average, LLaMA-3.1 8Baccuracy: 59.05 -> 60.79Improved
apply SRR with 2-bit QuIP# quantizerQERA-exact · WikiText2 LLaMA-3.1 8Bperplexity: 14.84 -> 13.73Improved
apply SRR with 3-bit GPTQ quantizerQERA-exact · WikiText2 LLaMA-2 7Bperplexity: 10.06 -> 9.98Improved
use SRR initialization for QPEFT instead of QERARoBERTa-base 2-bit QPEFT · GLUE averageaverage score: 72.51 -> 78.43Improved
use SRR initialization for QPEFT instead of QERALLaMA-3.1 8B 2-bit QPEFT · GSM8Kaccuracy: 25.11 -> 26.87Improved
use SRR with 4-bit QPEFT on RTE-sized small gainsLLaMA-3.1 8B 4-bit QPEFT · GSM8Kaccuracy: 54.20 -> 54.34 vs QLoRANo clear effect
no gradient attenuation on preserved directions (gamma=1 vs 0.1) (ablation)SRR QPEFT, 2-bit · GLUE averageaverage score: 73.38 -> 78.43 when adding gamma=0.1Improved
freeze preserved directions entirely (gamma=0) vs gamma=0.1 (ablation)SRR QPEFT, 2-bit · GLUE averageaverage score: 71.10 -> 78.43 when using gamma=0.1Improved
use SGP instead of fixed gamma=0.1 (ablation)SRR QPEFT, 2-bit · GLUE averageaverage score: 78.43 -> 78.88No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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