mlsanityData to 14 Oct 2026

Towards Understanding Distilled Reasoning Models: A Representational Approach

arXiv 2503.03730 ↗First posted 5 Mar 2025cs.LGcited by 26

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

also in 1 more
Abstract (arXiv)

In this paper, we investigate how model distillation impacts the development of reasoning features in large language models (LLMs). To explore this, we train a crosscoder on Qwen-series models and their fine-tuned variants. Our results suggest that the crosscoder learns features corresponding to various types of reasoning, including self-reflection and computation verification. Moreover, we observe that distilled models contain unique reasoning feature directions, which could be used to steer the model into over-thinking or incisive-thinking mode. In particular, we perform analysis on four specific reasoning categories: (a) self-reflection, (b) deductive reasoning, (c) alternative reasoning, and (d) contrastive reasoning. Finally, we examine the changes in feature geometry resulting from the distillation process and find indications that larger distilled models may develop more structured representations, which correlate with enhanced distillation performance. By providing insights into how distillation modifies the model, our study contributes to enhancing the transparency and reliability of AI systems.

Key facts. Cited by 26 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 29 of its 38 references that are in the map. First posted on arXiv 5 Mar 2025.

Builds on

29 of 38 references.

References · 29

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  3. 2020Scaling Laws for Neural Language Modelscited by 6,586
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2022Training Compute-Optimal Large Language Modelscited by 3,019
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  7. 2024OpenAI o1 System Cardcited by 1,244
  8. 2023Steering Language Models With Activation Engineeringcited by 922
  9. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  10. 2024Scaling and evaluating sparse autoencoderscited by 553
  11. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 394
  12. 2023Language Models Represent Space and Timecited by 316
17 more
  1. 2023Function Vectors in Large Language Modelscited by 308
  2. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 300
  3. 2018Model compression via distillation and quantizationcited by 289
  4. 2024A Survey on Knowledge Distillation of Large Language Modelscited by 284
  5. 2024Not All Language Model Features Are One-Dimensionally Linearcited by 212
  6. 2021Implicit Representations of Meaning in Neural Language Modelscited by 180
  7. 2016Long Short-Term Memory-Networks for Machine Readingcited by 161
  8. 2024The Geometry of Categorical and Hierarchical Concepts in Large Language Modelscited by 159
  9. 2022Omnigrok: Grokking Beyond Algorithmic Datacited by 146
  10. 2025Language Models Use Trigonometry to Do Additioncited by 70
  11. 2023Improving Activation Steering in Language Models with Mean-Centringcited by 68
  12. 2024Extending Activation Steering to Broad Skills and Multiple Behaviourscited by 40
  13. 2025Transcoders Beat Sparse Autoencoders for Interpretabilitycited by 25
  14. 2023Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Modelscited by 21
  15. 2024Opening the AI black box: program synthesis via mechanistic interpretabilitycited by 20
  16. 2024Investigating Representation Universality: Case Study on Genealogical Representationscited by 4
  17. 2025Harmonic Loss Trains Interpretable AI Modelscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 18; 2026: 8.

26 papers.

Sort:

By area:

  1. 2025Speculative Thinking: Enhancing Small-Model Reasoning with Large Model Guidance at Inference Timecited by 48
  2. 2025Mitigating Overthinking in Large Reasoning Models via Manifold Steeringcited by 33
  3. 2025Base Models Know How to Reason, Thinking Models Learn Whencited by 27
  4. 2025Emergent Coordination in Multi-Agent Language Modelscited by 23
  5. 2025Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilitiescited by 20
  6. 2025On Reasoning Strength Planning in Large Reasoning Modelscited by 13
  7. 2025Reasoning or Retrieval? A Study of Answer Attribution on Large Reasoning Modelscited by 10
  8. 2025Lexical Hints of Accuracy in LLM Reasoning Chainscited by 9
  9. 2025Demystifying Hybrid Thinking: Can LLMs Truly Switch Between Think and No-Think?cited by 9
  10. 2026Large Reasoning Models Are (Not Yet) Multilingual Latent Reasonerscited by 9
  11. 2025Crosscoding Through Time: Tracking Emergence & Consolidation Of Linguistic Representations Throughout LLM Pretrainingcited by 6
  12. 2025Internal states before wait modulate reasoning patternscited by 6
14 more
  1. 2026Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failurescited by 6
  2. 2025Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncationcited by 5
  3. 2025CodeReasoner: Enhancing the Code Reasoning Ability with Reinforcement Learningcited by 4
  4. 2025Rank-1 LoRAs Encode Interpretable Reasoning Signalscited by 4
  5. 2025Latent Concept Disentanglement in Transformer-based Language Modelscited by 3
  6. 2026Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Modelscited by 3
  7. 2025Distractor Injection Attacks on Large Reasoning Models: Characterization and Defensecited by 2
  8. 2026Transcoder Adapters for Reasoning-Model Diffingcited by 1
  9. 2026Fine-Tuning Small Reasoning Models for Quantum Field Theorycited by 1
  10. 2026Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoderscited by 1
  11. 2025Distilled Circuits: A Mechanistic Study of Internal Restructuring in Knowledge Distillationcited by 0
  12. 2025Enhancing Reasoning Capabilities in SLMs with Reward Guided Dataset Distillationcited by 0
  13. 2026fmxcoders: Factorized Masked Crosscoders for Cross-Layer Feature Discoverycited by 0
  14. 2026A Survey on the Linear Representation Hypothesiscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Small Models Struggle to Learn from Strong Reasoners
  2. 2023Mixed Distillation Helps Smaller Language Model Better Reasoning
  3. 2024Improving Mathematical Reasoning Capabilities of Small Language Models via Feedback-Driven Distillation
  4. 2025Not All Correct Answers Are Equal: Why Your Distillation Source Matters
  5. 2025Thinking with DistilQwen: A Tale of Four Distilled Reasoning and Reward Model Series
  6. 2025Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning
  7. 2026MI-Distillation: Selecting from Model-Interpolated Instruct-Reasoning Data Spectrum for Chain-of-Thought Distillation
  8. 2026Masked Distillation: Internalizing the Chain-of-Thought in Language Models