Research map / Large language models / Instruction and fine-tuning
Neuron-Aware Data Selection in Instruction Tuning for Large Language Models
Xin Chen, Junchao Wu, Shu Yang, Runzhe Zhan, Zeyu Wu, Min Yang, Shujian Huang, Lidia S. Chao, Derek F. Wong
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 changed | Where | Measured | Effect |
|---|---|---|---|
| select 10% of Alpaca-GPT4 with NAIT (all features) instead of training on full data | LLaMA-2-7b · 9-benchmark average | average score: 36.03 -> 37.20 (+3.24%) | Improved |
| select data using GSM activation features | LLaMA-2-7b · 9-benchmark average | average score: 36.03 -> 37.70 (+4.65%) | Improved |
| use NAIT selection instead of AlpaGasus | LLaMA-2-7b · 9-benchmark average | average score: 35.18 -> 37.20 | Improved |
| use NAIT (all features) instead of SelectIT | LLaMA-2-7b · 9-benchmark average | average score: 37.16 -> 37.20 | No clear effect |
| choose top 10% highest-alignment samples instead of random (ablation) | LLaMA-2-7b · 5-task average | average score: 34.04 -> 35.18 | Improved |
| choose lowest-alignment samples instead of random (ablation) | LLaMA-2-7b · 5-task average | average score: 34.04 -> 28.27 | Hurt |
| apply NAIT selection on LLaMA-2-13b (ablation) | LLaMA-2-13b · 5-task average | average score: 40.76 -> 42.53 vs random | Improved |
| apply NAIT selection on Mistral-7b (ablation) | Mistral-7b · 5-task average | average score: 37.69 -> 41.43 vs random | Improved |
| apply NAIT selection on Llama-3-8b (ablation) | Llama-3-8b · 5-task average | average score: 52.10 -> 57.63 vs random | Improved |
| apply NAIT selection on Qwen-2.5-7b (ablation) | Qwen-2.5-7b · 5-task average | average score: 73.39 -> 75.20 vs random | Improved |
| increase selected fraction beyond 30% up to 100% (ablation) | NAIT · Alpaca-GPT4 | performance | Hurt |
| increase number of in-domain examples (ablation) | NAIT · GSM and TydiQA | score: peaks 18.35 (GSM) and 47.57 (TydiQA) at 4096 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SelectIT — SelectIT: Selective Instruction Tuning for LLMs via Uncertainty-Aware Self-Reflection (NeurIPS 2024) · better on 3 of 4 shared results
- LIMA — LIMA: Less Is More for Alignment (NeurIPS 2023) · better on 1 of 1 shared result
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
- Task-Specific Data Selection for Instruction Tuning via Monosemantic Neuronal Activations (NeurIPS 2025 · similarity 0.92)
- SelectIT: Selective Instruction Tuning for LLMs via Uncertainty-Aware Self-Reflection (NeurIPS 2024 · similarity 0.90)
- LESS: Selecting Influential Data for Targeted Instruction Tuning (ICML 2024 · similarity 0.90)
- Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data (NeurIPS 2025 · similarity 0.88)
- What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuning (ICLR 2024 · similarity 0.88)
- MAmmoTH2: Scaling Instructions from the Web (NeurIPS 2024 · similarity 0.88)
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.
