mlsanityData to 7 Oct 2026

Entropy and Attention Dynamics in Small Language Models: A Trace-Level Structural Analysis on the TruthfulQA Benchmark

arXiv 2604.03589 ↗First posted 4 Apr 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Small language models (SLMs) have been increasingly deployed in edge devices and other resource-constrained settings. However, these models make confident mispredictions and produce unstable output, making them risky for factual and decision-critical tasks. Current evaluation methodology relies on final accuracy or hallucination rates without explaining how internal model behavior affects outputs. Specifically, how entropy evolves during decoding, how attention is distributed across layers, and how hidden representations contribute to uncertainty, logical inconsistencies, and misinformation propagation are often overlooked. Consequently, this study introduces a trace-level analysis of entropy and attention dynamics in SLMs evaluated with the TruthfulQA dataset. Four models with parameter ranges of 1B-1.7B parameters were examined via token-level output entropy, attention entropy, head dispersion, and hidden-state representation. The results reflect three model classifications by entropy patterns. Deterministic models (DeepSeek-1.5B and LLaMA-1B): output entropy decreases over time. Exploratory models (Gemma-1B): with increasing entropy, and balanced models (Qwen-1.7B): have moderate and stable entropy. Also, each group has distinctively different hidden-state movement and attention dispersion patterns. The analysis demonstrates that truthfulness in SLMs emerges from structured entropy and attention dynamics. Monitoring and optimizing these internal uncertainty patterns can guide the design of a more reliable, hallucination-aware, and application-specific edge SLMs.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 26 of its 47 references that are in the map. First posted on arXiv 4 Apr 2026.

Builds on

26 of 47 references.

References · 26

  1. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 844
  2. 2021Measuring and Improving Consistency in Pretrained Language Modelscited by 361
  3. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 293
  4. 2023A Survey of Confidence Estimation and Calibration in Large Language Modelscited by 253
  5. 2019Analyzing the Structure of Attention in a Transformer Language Modelcited by 241
  6. 2023Domain Specialization as the Key to Make Large Language Models Disruptive: A Comprehensive Surveycited by 134
  7. 2023Factuality Challenges in the Era of Large Language Modelscited by 125
  8. 2024Mobile Edge Intelligence for Large Language Models: A Contemporary Surveycited by 125
  9. 2024A Survey on Uncertainty Quantification of Large Language Models: Taxonomy, Open Research Challenges, and Future Directionscited by 115
  10. 2025Uncertainty Quantification and Confidence Calibration in Large Language Models: A Surveycited by 93
  11. 2024Calibrating Language Models with Adaptive Temperature Scalingcited by 57
  12. 2019Calibration, Entropy Rates, and Memory in Language Modelscited by 32
14 more
  1. 2025Vision-Language Models for Edge Networks: A Comprehensive Surveycited by 29
  2. 2024Attend First, Consolidate Later: On the Importance of Attention in Different LLM Layerscited by 25
  3. 2025Do Language Models Mirror Human Confidence? Exploring Psychological Insights to Address Overconfidence in LLMscited by 19
  4. 2025Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Stylescited by 12
  5. 2025Are the Hidden States Hiding Something? Testing the Limits of Factuality-Encoding Capabilities in LLMscited by 11
  6. 2019Multi-Head Attention with Diversity for Learning Grounded Multilingual Multimodal Representationscited by 10
  7. 2025Large Language-Geometry Model: When LLM meets Equivariancecited by 10
  8. 2026EDIS: Diagnosing LLM Reasoning via Entropy Dynamicscited by 10
  9. 2025Simple Yet Effective: An Information-Theoretic Approach to Multi-LLM Uncertainty Quantificationcited by 6
  10. 2025Cognitive Mirrors: Exploring the Diverse Functional Roles of Attention Heads in LLM Reasoningcited by 5
  11. 2022Normal Transformer: Extracting Surface Geometry from LiDAR Points Enhanced by Visual Semanticscited by 2
  12. 2024Detecting Errors through Ensembling Prompts (DEEP): An End-to-End LLM Framework for Detecting Factual Errorscited by 2
  13. 2025Exploring specialization and sensitivity of convolutional neural networks in the context of simultaneous image augmentationscited by 2
  14. 2025Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attentioncited by 2

Built on by

Not cited yet.