AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

IDEAL: Data Equilibrium Adaptation for Multi-Capability Language Model Alignment

Chenlin Ming, Chendi Qu, Qizhi Pei, Zhuoshi Pan, Yu Li, Xiaoming Duan, Lijun Wu, Conghui He

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 5 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 IDEAL gradient-based domain re-weighting instead of Joint SFT uniform mixtureLlama3.1-8B SFT · Overall average, 4 benchmarks, 1 epochaverage score: 54.79 -> 57.87Improved
use IDEAL re-weighting instead of Joint SFTLlama3.1-8B SFT · Overall average, 4 benchmarks, 3 epochsaverage score: 55.35 -> 59.23Improved
use IDEAL re-weighting instead of Joint SFTLlama3.1-8B SFT · HumanEval, 1 epochpass@1: 41.26 -> 50.61Improved
use IDEAL re-weighting on balanced 5-domain mixtureLlama3.1-8B SFT · Extended 8-benchmark averageaverage score: 54.03 -> 55.20Improved
use IDEAL re-weighting on balanced 5-domain mixtureLlama3.1-8B SFT · MATH (extended)accuracy: 30.94 -> 30.20Hurt
increase joint SFT data from 25k to full 66k (ablation)Joint SFT · Extended 8-benchmark averageaverage score: 54.03 -> 53.94No clear effect
upsample DoReMi data to IDEAL's volume (ablation)DoReMi D2 · Overall average, 1 epochaverage score: 55.17 -> 55.36No clear effect
upsample DOGE data to IDEAL's volume (ablation)DOGE D2 · Overall average, 1 epochaverage score: 53.37 -> 53.81No clear effect
set scale m to 0.15 rather than 0.1 or 0.3 (ablation)IDEAL · Sensitivity of m, D1 1 epochaverage scoreImproved
run second IDEAL iteration (D1 to D2) (ablation)IDEAL · Overall average, 1 epochaverage score: 56.28 -> 57.87Improved
train 3 epochs instead of 1 epoch (ablation)IDEAL D2 · HumanEvalpass@1: 50.61 -> 46.34Hurt

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