AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Arbitrary-Order Block SignSGD for Memory-Efficient LLM Fine-Tuning

Yijie Zhou, Shi Pu

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging

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 depth-biased block selection (vs updating all blocks as in BAdam)ABSignSGD · OpenMathInstruct-2 runtimeruntime (h): 3.32 -> 2.66Improved
replace Adam with SignSGD as the base optimizer under block updates (ablation)block-coordinate optimizer · Qwen3-1.7B OpenMathInstruct-2training loss: BAdam impaired vs Adam; ABSignSGD fastestImproved
use SignSGD instead of SGD in block-coordinate updates (ablation)BlockSGD · Qwen3-1.7B OpenMathInstruct-2training loss: faster iteration-wise convergence than SGDImproved
vary block selection rule (depth-biased, deep-to-shallow, uniform random) (ablation)ABSignSGD · Qwen3-1.7B OpenMathInstruct-2training loss: similar convergence across DS, UR, DBNo clear effect
add majority-vote aggregation over 4 agents (ablation)ABSignSGD · Qwen3-8B OpenMathInstruct-2training loss: tracks non-MV baseline closelyNo clear effect
scale MV agents from 1 to 32 (ablation)ABSignSGD-MV · Qwen3-8B OpenMathInstruct-2training loss: global batch size 4 to 128No clear effect
use ABSignSGD instead of best baseline (BAdam) for math fine-tuningQwen3-8B · math benchmarks averageaccuracy: 70 -> 76Improved
use ABSignSGD instead of BAdam for instruction tuningQwen3-8B · MT-Bench averagescore: 5.72 -> 6.18Improved
use ABSignSGD instead of Apollo for instruction tuningLlama3-8B · MT-Bench averagescore: 5.67 -> 5.99Improved

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