AI Paper Atlas

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

Mining Useful General Data for Low-Resource Domain Adaptation

Pingjie Wang, Hongcheng Liu, Yusheng Liao, Ziqing Fan, Yaxin Du, Shuo Tang, Yanfeng Wang, Yu Wang

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
fine-tune on only 1K domain dataLlama3-8B-Instruct · ContractNLIaccuracy: 59.9 -> 41.1Hurt
fine-tune on only 1K domain dataQwen3-8B · FPBaccuracy: 80.1 -> 72.2Hurt
add 9K randomly selected auxiliary dataLlama3-8B-Instruct · average of 8 tasksaccuracy: 68.7 -> 74.3Improved
add 9K randomly selected auxiliary dataQwen3-8B · average of 8 tasksaccuracy: 74.4 -> 72.9Hurt
select auxiliary data with NTK-Selector instead of randomLlama3-8B-Instruct · average of 8 tasksaccuracy: 74.3 -> 76.6Improved
select auxiliary data with NTK-Selector instead of DSIRQwen3-8B · average of 8 tasksaccuracy: 77.7 -> 78.6Improved
select auxiliary data with LESSQwen3-8B · FPBaccuracy: 80.1 -> 60.3Hurt
increase total data from x1 to x5 with NTK-Selector auxiliary data (ablation)NTK-Selector · TFNSaccuracy: 69.3 -> 71.9Improved
increase total data from x10 to x20 (ablation)NTK-Selector · TFNSaccuracy: 73.3 -> 73.3No clear effect
increase pre-selection size from 2N to 16N (ablation)NTK-Selector · MMLU-Medaccuracy: 73.3 -> 74.4Improved
increase pre-selection size from 2N to 16N (ablation)NTK-Selector · FPBaccuracy: 84.5 -> 86.9Improved
increase pre-selection size from 4N to 16N (ablation)NTK-Selector · RSDDaccuracy: 96.5 -> 95.9No clear effect

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