Research map / Large language models / Instruction and fine-tuning
Utility-Diversity Aware Online Batch Selection for LLM Supervised Fine-tuning
Heming Zou, Yixiu Mao, Yun Qu, Qi Wang, Xiangyang Ji
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 |
|---|---|---|---|
| Use UDS (nuclear norm plus diversity distance) instead of GREATS for online batch selection | Qwen-2.5-7B · MMLU | average accuracy (%): 58.19 -> 63.34 | Improved |
| Use UDS instead of full-dataset SFT | Qwen-2.5-7B · MMLU | average accuracy (%): 55.32 -> 63.34 | Improved |
| Use UDS instead of full-dataset SFT | Qwen-2.5-7B · MMLU training throughput | samples per second: 2.27 -> 3.41 | Improved |
| Use UDS instead of full-dataset SFT | Qwen-2.5-7B · ScienceQA | average accuracy (%): 94.56 -> 95.19 | Improved |
| Add nuclear norm score to random selection (ablation) | Random selection baseline · MMLU | average accuracy (%): 54.26 -> 58.35 | Improved |
| Add diversity distance to random selection (ablation) | Random selection baseline · MMLU | average accuracy (%): 54.26 -> 57.75 | Improved |
| Combine diversity distance with nuclear norm (ablation) | UDS nuclear-norm-only · MMLU | average accuracy (%): 58.35 -> 63.34 | Improved |
| Combine diversity distance with nuclear norm (ablation) | UDS nuclear-norm-only · HumanEval | Pass@1 (%): 44.18 -> 46.28 | Improved |
| Select gradient-norm-heavy samples (MaxGrad) instead of random | online batch selection · MMLU training throughput | samples per second: Qwen-2.5-7B: 0.31 vs 2.27 full-dataset | Hurt |
| Select max-loss samples (MaxLoss) instead of random | online batch selection · MMLU | average accuracy (%): 54.26 -> 54.51 | No clear effect |
| Increase selected samples K per batch beyond 4 (ablation) | UDS · MMLU (Llama-3.1-8B) | average accuracy (%) | Hurt |
| Use UDS at K=4 instead of full-dataset fine-tuning | Llama-3.1-8B · MMLU | average accuracy (%) | Improved |
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.
- GREATS — GREATS: Online Selection of High-Quality Data for LLM Training in Every Iteration (NeurIPS 2024) · better on 8 of 8 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
- Data Mixing Optimization for Supervised Fine-Tuning of Large Language Models (ICML 2025 · similarity 0.91)
- FisherSFT: Data-Efficient Supervised Fine-Tuning of Language Models Using Information Gain (ICML 2025 · similarity 0.89)
- SmallToLarge (S2L): Scalable Data Selection for Fine-tuning Large Language Models by Summarizing Training Trajectories of Small Models (NeurIPS 2024 · similarity 0.89)
- One Algorithm, Two Goals: Dual Scoring for Parameter and Data Selection in LLM Fine-Tuning (NeurIPS 2026 · similarity 0.89)
- Towards Scalable Data Diversification for Language Model Pretraining via Leverage Score Sampling (NeurIPS 2026 · similarity 0.89)
- Get more for less: Principled Data Selection for Warming Up Fine-Tuning in LLMs (ICLR 2024 · similarity 0.89)
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.
