mlsanityData to 7 Oct 2026

Find Your Optimal Teacher: Personalized Data Synthesis via Router-Guided Multi-Teacher Distillation

arXiv 2510.10925 ↗First posted 13 Oct 2025cs.LG cs.CLcited by 15

On the map: Synthetic data and self-improvement for LLMs → Code and instruction data synthesis

Abstract (arXiv)

Training student models on synthetic data generated by strong teacher models is a promising way to distilling the capabilities of teachers. However, recent studies show that stronger models are not always optimal teachers, revealing a mismatch between teacher outputs and student learnability. To address this issue, we propose PerSyn (Personalized data Synthesis), a novel synthesis strategy that operates under a new ``Route then Generate'' paradigm to create data tailored to each student model, enabling it to learn more effectively. Specifically, PerSyn first assigns each prompt to its optimal teacher via a query-level router that jointly considers student learnability and teacher response quality. Each teacher then synthesizes data only for its assigned prompts, making the process more efficient than the conventional ``Generate then Select'' paradigm, where all teachers must generate parallel responses for the entire prompt set before constructing the final dataset. Extensive experiments across different model families and scales demonstrate that PerSyn consistently achieves superior or comparable performance to all baselines in instruct tuning and math reasoning settings. Further analysis verifies the effectiveness of PerSyn and offers extra insights to propel future research.

Key facts. Cited by 15 papers in the map (data through 7 Oct 2026). Placed in Code and instruction data synthesis, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 31 of its 52 references that are in the map. First posted on arXiv 13 Oct 2025.

Builds on

31 of 52 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025Improving Natural Language Understanding for LLMs via Large-Scale Instruction Synthesis6 Feb

Other references · 30

  1. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  2. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  3. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
  4. 2023Code Llama: Open Foundation Models for Codecited by 2,105
  5. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,883
  6. 2024LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Modelscited by 1,442
  7. 2021Are NLP Models really able to Solve Simple Math Word Problems?cited by 861
  8. 2025DeepSeek-Prover-V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decompositioncited by 270
  9. 2024On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Surveycited by 240
  10. 2024Large Language Models for Data Annotation and Synthesis: A Surveycited by 201
  11. 2022Distilling Reasoning Capabilities into Smaller Language Modelscited by 155
  12. 2025Small Models Struggle to Learn from Strong Reasonerscited by 109
18 more
  1. 2025Beyond Pass@1: Self-Play with Variational Problem Synthesis Sustains RLVRcited by 53
  2. 2025SwS: Self-aware Weakness-driven Problem Synthesis in Reinforcement Learning for LLM Reasoningcited by 39
  3. 2024Evaluating Language Models as Synthetic Data Generatorscited by 32
  4. 2025TreeReview: A Dynamic Tree of Questions Framework for Deep and Efficient LLM-based Scientific Peer Reviewcited by 31
  5. 2024SMoA: Improving Multi-agent Large Language Models with Sparse Mixture-of-Agentscited by 30
  6. 2025Chain-of-Reasoning: Towards Unified Mathematical Reasoning in Large Language Models via a Multi-Paradigm Perspectivecited by 29
  7. 2024SimRAG: Self-Improving Retrieval-Augmented Generation for Adapting Large Language Models to Specialized Domainscited by 28
  8. 2024ToolFlow: Boosting LLM Tool-Calling Through Natural and Coherent Dialogue Synthesiscited by 27
  9. 2024PromptKD: Distilling Student-Friendly Knowledge for Generative Language Models via Prompt Tuningcited by 24
  10. 2024BPO: Towards Balanced Preference Optimization between Knowledge Breadth and Depth in Alignmentcited by 20
  11. 2024Initialization using Update Approximation is a Silver Bullet for Extremely Efficient Low-Rank Fine-Tuningcited by 18
  12. 2024Genetic Instruct: Scaling up Synthetic Generation of Coding Instructions for Large Language Modelscited by 17
  13. 2024Learning from Committee: Reasoning Distillation from a Mixture of Teachers with Peer-Reviewcited by 14
  14. 2024Multilingual Large Language Models and Curse of Multilingualitycited by 13
  15. 2024Task Oriented In-Domain Data Augmentationcited by 6
  16. 2025GuiLoMo: Allocating Expert Number and Rank for LoRA-MoE via Bilevel Optimization with GuidedSelection Vectorscited by 6
  17. 2024Fostering Natural Conversation in Large Language Models with NICO: a Natural Interactive COnversation datasetcited by 4
  18. 2025Multiple Choice Learning of Low-Rank Adapters for Language Modelingcited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 15.

15 papers.

Sort:

By area:

  1. 2026ToolPRMBench: Evaluating and Advancing Process Reward Models for Tool-using Agentscited by 13
  2. 2026Sustainable Hybrid Document-Routed Retrieval for Financial RAG: Resolving the Robustness-Precision Trade-offcited by 13
  3. 2026Bridging Reasoning Trajectories in On-Policy Distillation via Near-Future Guidancecited by 12
  4. 2026Cornerstones or Stumbling Blocks? Deciphering the Rock Tokens in On-Policy Distillationcited by 11
  5. 2026Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Modelscited by 7
  6. 2026Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertaintycited by 6
  7. 2026Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidancecited by 5
  8. 2026MIS-HCC: Hierarchical Channel Clustering for Efficient Medical Image Segmentationcited by 4
  9. 2026MUSE: Resolving Manifold Misalignment in Visual Tokenization via Topological Orthogonalitycited by 3
  10. 2026Polyglot Teachers: Evaluating Language Models for Multilingual Synthetic Data Generationcited by 1
  11. 2026When the Strongest Teacher Is Not the Best Teacher: Student-Centric Answer Selectioncited by 1
  12. 2026IMAGINE: Adaptive Schema-Imagery Enhanced Composition for Composed Video Retrievalcited by 1
3 more
  1. 2026Energy-Efficient On-Device RAG on a Mobile NPU: System Design and Benchmark on Snapdragon X Elitecited by 1
  2. 2026MOSAIC: Orchestrating Collaborative Knowledge Tracing with Hierarchical Semantic Alignmentcited by 0
  3. 2026Hallucination Self-Play: Bootstrapping Reinforced Detector via Evolved Generatorcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Montessori-Instruct: Generate Influential Training Data Tailored for Student Learning
  2. 2026How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data
  3. 2024Aligning Teacher with Student Preferences for Tailored Training Data Generation
  4. 2026PADD: Path-Aligned Decompression Distillation for Non-Router Teacher to Guide MoE Student Learning
  5. 2025The Signal is in the Steps: Local Scoring for Reasoning Data Selection
  6. 2024Rethinking Data Synthesis: A Teacher Model Training Recipe with Interpretation
  7. 2025Distillation Scaling Laws
  8. 2026Learning from Teacher Continuations at Student States