mlsanityData to 7 Oct 2026

Diversity-driven Data Selection for Language Model Tuning through Sparse Autoencoder

arXiv 2502.14050 ↗First posted 19 Feb 2025cs.CL cs.AI cs.LGcited by 11

Not on the map. Why?

Abstract (arXiv)

Instruction tuning data are often quantity-saturated due to the large volume of data collection and fast model iteration, leaving data selection important but underexplored. Existing quality-driven data selection methods, such as LIMA (NeurIPS 2023 citepzhou2024lima) and AlpaGasus (ICLR 2024 citepchenalpagasus) generally ignore the equal importance of data diversity and complexity. In this work, we aim to design a diversity-aware data selection strategy and creatively propose using sparse autoencoders (SAEs) to tackle the challenge of data diversity measure. In addition, SAEs can also provide more interpretability of model behavior and explain, e.g., the surprising effectiveness of selecting the longest response (ICML 2024 citepzhaolong). Using effective data selection, we experimentally prove that models trained on our selected data can outperform other methods in terms of model capabilities, reduce training cost, and potentially gain more control over model behaviors. We prove that SAEs can serve as a good alternative to diversity measure and design our method to be scalable for potential industrial large-scale pruning, and we will also release our trained SAEs for use by the broader community.

Key facts. Cited by 11 papers in the map (data through 7 Oct 2026). Builds on 72 of its 86 references that are in the map. First posted on arXiv 19 Feb 2025.

Builds on

72 of 86 references.

References · 72

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2022Training language models to follow instructions with human feedbackcited by 15,712
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2024The Llama 3 Herd of Modelscited by 14,192
  5. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  6. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  7. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  8. 2020Scaling Laws for Neural Language Modelscited by 6,411
  9. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  10. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  11. 2023Qwen Technical Reportcited by 3,273
  12. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
38 more
  1. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
  2. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,674
  3. 2024Qwen2 Technical Reportcited by 2,076
  4. 2021Multitask Prompted Training Enables Zero-Shot Task Generalizationcited by 1,462
  5. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  6. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
  7. 2023WizardLM: Empowering large pre-trained language models to follow complex instructionscited by 1,116
  8. 2022Toy Models of Superpositioncited by 928
  9. 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 865
  10. 2023Enhancing Chat Language Models by Scaling High-quality Instructional Conversationscited by 825
  11. 2023The Flan Collection: Designing Data and Methods for Effective Instruction Tuningcited by 734
  12. 2024Scaling and evaluating sparse autoencoderscited by 530
  13. 2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 529
  14. 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 514
  15. 2023UltraFeedback: Boosting Language Models with Scaled AI Feedbackcited by 394
  16. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
  17. 2023What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuningcited by 354
  18. 2023Finding Neurons in a Haystack: Case Studies with Sparse Probingcited by 328
  19. 2023Data Selection for Language Models via Importance Resamplingcited by 322
  20. 2023From Quantity to Quality: Boosting LLM Performance with Self-Guided Data Selection for Instruction Tuningcited by 301
  21. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 283
  22. 2020Understanding the Role of Individual Units in a Deep Neural Networkcited by 254
  23. 2022Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learntcited by 233
  24. 2024A Survey on Data Selection for Language Modelscited by 230
  25. 2023Camels in a Changing Climate: Enhancing LM Adaptation with Tulu 2cited by 227
  26. 2013k-Sparse Autoencoderscited by 209
  27. 2024Data Engineering for Scaling Language Models to 128K Contextcited by 183
  28. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 171
  29. 2023D4: Improving LLM Pretraining via Document De-Duplication and Diversificationcited by 146
  30. 2024Superfiltering: Weak-to-Strong Data Filtering for Fast Instruction-Tuningcited by 134
  31. 2023#InsTag: Instruction Tagging for Analyzing Supervised Fine-tuning of Large Language Modelscited by 108
  32. 2023MoDS: Model-oriented Data Selection for Instruction Tuningcited by 107
  33. 2021Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factorscited by 106
  34. 2024Automatically Interpreting Millions of Features in Large Language Modelscited by 106
  35. 2023NEFTune: Noisy Embeddings Improve Instruction Finetuningcited by 99
  36. 2024Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuningcited by 85
  37. 2023Data Diversity Matters for Robust Instruction Tuningcited by 76
  38. 2024A Survey on Data Selection for LLM Instruction Tuningcited by 76

50 most cited of 72 All 72 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 8; 2026: 3.

11 papers.

  1. 2025Measuring Data Diversity for Instruction Tuning: A Systematic Analysis and A Reliable Metriccited by 24
  2. 2025Verifying Chain-of-Thought Reasoning via Its Computational Graphcited by 20
  3. 2025Select2Reason: Efficient Instruction-Tuning Data Selection for Long-CoT Reasoningcited by 7
  4. 2025Task-Specific Data Selection for Instruction Tuning via Monosemantic Neuronal Activationscited by 4
  5. 2025Computational Budget Should Be Considered in Data Selectioncited by 2
  6. 2026Rethinking Representativeness and Diversity in Dynamic Data Selectioncited by 2
  7. 2025Text2Cypher: Data Pruning using Hard Example Selectioncited by 1
  8. 2025Learning from the Best, Differently: A Diversity-Driven Rethinking on Data Selectioncited by 1
  9. 2026Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoderscited by 1
  10. 2025Exploring the Heterogeneity of Tabular Data: A Diversity-aware Data Generator via LLMscited by 0
  11. 2026Towards Scalable Data Diversification for Language Model Pretraining via Leverage Score Samplingcited by 0