AI Paper Atlas

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

Decomposing the Basic Abilities of Large Language Models: Mitigating Cross-Task Interference in Multi-Task Instruct-Tuning

Bing Wang, Ximing Li, Changchun Li, Jinjin Chi, Gang Niu, Masashi Sugiyama

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 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
add BADIT (SVD-based ability decomposition plus orthogonal grouping) over GainLoRAmulti-task LoRA-MoE · SuperNI, mixed trainingROUGE: 54.33 -> 55.87 (Qwen3-8B)Improved
add BADITLoRAMoE on Llama3-8B · SuperNI, sequential trainingROUGE: 45.47 -> 48.83Improved
add BADITGainLoRA on Qwen3-8B · SuperNI, sequential trainingForget Rate: 8.96 -> 6.95Improved
add SVD-based basic ability decomposition initialization (BAD) (ablation)BADIT · SuperNI sequential, Qwen3-8BROUGE: 50.07 -> 50.86Improved
add dynamically orthogonal grouping (DOG) (ablation)BADIT · SuperNI sequential, Qwen3-8BROUGE: 49.70 -> 50.86Improved
add both BAD and DOG (ablation)LoRAMoE · SuperNI sequential, Llama3-8BROUGE: 45.47 -> 48.83Improved
add DOG (ablation)BADIT · SuperNI mixed, Gemma2-9BROUGE: 56.03 -> 57.35Improved
add dynamic regrouping of rank-1 componentsLoRAMoE on Qwen3-8B · gradient anglesinter-expert gradient angle: near 90 degrees maintained throughout trainingImproved
add BADIT over LoRAMoELoRAMoE · training timerelative training time: 1x -> 1.22xHurt
add SVD-based initialization without DOG (ablation)LoRAMoE · training timerelative training time: 1x -> 1.00xNo clear effect
use PiSSA instead of LoRAmulti-task LoRA · SuperNI mixed, Qwen3-8BROUGE: 54.22 -> 53.57Hurt

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