AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Calibrating and Rotating: A Unified Framework for Weight Conditioning in PEFT

Da Chang, Peng Xue, Yu Li, Yongxiang Liu, Pengxiang Xu, Shixun Zhang

AAAI 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 1 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 SORA (pre-diagonal calibration plus orthogonal rotation) over LoRALoRA · GLUE avg, DeBERTaV3-baseaverage score: 88.57 -> 89.19Improved
add SORA over DoRADoRA · Math reasoning, Gemma-7Baverage accuracy: 78.23 -> 78.96Improved
add Pre-Diag diagonal calibration on pre-trained weights over LoRA (ablation)LoRA · GLUE avg, DeBERTaV3-baseaverage score: 88.57 -> 88.98Improved
apply orthogonal rotation before the LoRA update (Pre-Ortho) (ablation)SORA · GLUE avg, DeBERTaV3-baseaverage GLUE score: 88.57 -> 87.76 (vs LoRA)Hurt
add Pre-Diag calibration before post-rotation (ablation)Post-Ortho · GLUE avg, DeBERTaV3-baseaverage GLUE score: 88.93 -> 89.23Improved
increase rotational rank rP beyond 1 (ablation)SORA · GLUE RTE, DeBERTaV3-baseaccuracy: 87.90 -> 89.53 at rP=16, training speed 2.17 -> 1.72Mixed
replace DoRA with Pre-Diag or SORA for speedDoRA · LLaMA3-8B training speedtraining speed: Pre-Diag +50.60%, SORA +36.96% faster than DoRAImproved
replace DoRA with Pre-Diag or SORA for inferenceDoRA · LLaMA3-8B inference speedinference speed: Pre-Diag +65.09%, SORA +51.27% faster than DoRAImproved
use Pre-Diag instead of LoRA at rank 4LoRA · Commonsense reasoning avg, LLaMA3-8B, rank 4average accuracy: 75.00 -> 74.27Hurt
use SORA instead of LoRA at rank 32LoRA · Commonsense reasoning avg, LLaMA3-8B, rank 32average accuracy: 77.63 -> 78.06Improved
DoRA update increases singular value entropy relative to LoRALoRA · GLUE, DeBERTaV3-baseSVD entropyImproved

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