mlsanityData to 14 Oct 2026

Calibrating Language Models with Adaptive Temperature Scaling

arXiv 2409.19817 ↗First posted 29 Sep 2024cs.LG cs.AI cs.CLcited by 58

On the map: LLM alignment / RLHF / preference optimisation → LLM preference optimization methods

Abstract (arXiv)

The effectiveness of large language models (LLMs) is not only measured by their ability to generate accurate outputs but also by their calibration-how well their confidence scores reflect the probability of their outputs being correct. While unsupervised pre-training has been shown to yield LLMs with well-calibrated conditional probabilities, recent studies have shown that after fine-tuning with reinforcement learning from human feedback (RLHF), the calibration of these models degrades significantly. In this work, we introduce Adaptive Temperature Scaling (ATS), a post-hoc calibration method that predicts a temperature scaling parameter for each token prediction. The predicted temperature values adapt based on token-level features and are fit over a standard supervised fine-tuning (SFT) dataset. The adaptive nature of ATS addresses the varying degrees of calibration shift that can occur after RLHF fine-tuning. ATS improves calibration by over 10-50% across three downstream natural language evaluation benchmarks compared to prior calibration methods and does not impede performance improvements from RLHF.

Key facts. Cited by 58 papers in the map (data through 14 Oct 2026). Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 27 of its 34 references that are in the map. First posted on arXiv 29 Sep 2024.

Builds on

27 of 34 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback24 May

Other references · 26

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2017On Calibration of Modern Neural Networkscited by 4,960
  5. 2023Qwen Technical Reportcited by 3,344
  6. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  7. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  8. 2022Language Models (Mostly) Know What They Knowcited by 1,856
  9. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,850
  10. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  11. 2019When Does Label Smoothing Help?cited by 988
  12. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
14 more
  1. 2018Deep Anomaly Detection with Outlier Exposurecited by 942
  2. 2023Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generationcited by 819
  3. 2022Teaching Models to Express Their Uncertainty in Wordscited by 751
  4. 2023Instruction Tuning with GPT-4cited by 686
  5. 2020Measuring Robustness to Natural Distribution Shifts in Image Classificationcited by 454
  6. 2019Verified Uncertainty Calibrationcited by 309
  7. 2019On Mixup Training: Improved Calibration and Predictive Uncertainty for Deep Neural Networkscited by 305
  8. 2020Reducing conversational agents' overconfidence through linguistic calibrationcited by 210
  9. 2022Uncertainty Quantification with Pre-trained Language Models: A Large-Scale Empirical Analysiscited by 104
  10. 2022Training OOD Detectors in their Natural Habitatscited by 86
  11. 2020Combining Ensembles and Data Augmentation can Harm your Calibrationcited by 44
  12. 2022On the Calibration of Pre-trained Language Models using Mixup Guided by Area Under the Margin and Saliencycited by 31
  13. 2023A Study on the Calibration of In-context Learningcited by 22
  14. 2023Conservative Prediction via Data-Driven Confidence Minimizationcited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 4; 2025: 22; 2026: 32.

58 papers.

Sort:

By area:

  1. 2024Knowledge Boundary of Large Language Models: A Surveycited by 40
  2. 2025Optimizing Temperature for Language Models with Multi-Sample Inferencecited by 33
  3. 2025Diverse Preference Learning for Capabilities and Alignmentcited by 25
  4. 2025From Calibration to Collaboration: LLM Uncertainty Quantification Should Be More Human-Centeredcited by 21
  5. 2025Know What You Don't Know: Uncertainty Calibration of Process Reward Modelscited by 21
  6. 2025Your Pre-trained LLM is Secretly an Unsupervised Confidence Calibratorcited by 20
  7. 2025Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuumcited by 18
  8. 2025Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence Awarenesscited by 14
  9. 2025Large Language Models Do Multi-Label Classification Differentlycited by 13
  10. 2025The Confidence Paradox: Can LLM Know When It's Wrongcited by 11
  11. 2024Adaptive Decoding via Latent Preference Optimizationcited by 10
  12. 2025Measuring Scalar Constructs in Social Science with LLMscited by 9
38 more
  1. 2025Calibrating LLM Judges: Linear Probes for Fast and Reliable Uncertainty Estimationcited by 9
  2. 2026BaseCal: Unsupervised Confidence Calibration via Base Model Signalscited by 8
  3. 2026Vision-Language Introspection: Mitigating Overconfident Hallucinations in MLLMs via Interpretable Bi-Causal Steeringcited by 8
  4. 2024Text-to-SQL Calibration: No Need to Ask -- Just Rescale Model Probabilitiescited by 7
  5. 2025Beyond Accuracy: The Role of Calibration in Self-Improving Large Language Modelscited by 7
  6. 2026NOVA: NOise-aware Verbal Confidence CAlibration for Robust Large Language Models in RAG Systemscited by 6
  7. 2026LLMs Should Express Uncertainty Explicitlycited by 6
  8. 2025Specialization after Generalization: Towards Understanding Test-Time Training in Foundation Modelscited by 5
  9. 2026Gradient-Stable Attention Heads Signal LLM Correctnesscited by 5
  10. 2026Instinct vs. Reflection: Unifying Token and Verbalized Confidence in Multimodal Large Modelscited by 4
  11. 2025Decoding Uncertainty: The Impact of Decoding Strategies for Uncertainty Estimation in Large Language Modelscited by 3
  12. 2025SCOPE: Stochastic and Counterbiased Option Placement for Evaluating Large Language Modelscited by 2
  13. 2025Less Diverse, Less Safe: The Indirect But Pervasive Risk of Test-Time Scaling in Large Language Modelscited by 2
  14. 2025SDA: Steering-Driven Distribution Alignment for Open LLMs without Fine-Tuningcited by 2
  15. 2026Improving Semantic Uncertainty Quantification in Language Model Question-Answering via Token-Level Temperature Scalingcited by 2
  16. 2026LLM Benchmark Datasets Should Be Contamination-Resistantcited by 2
  17. 2026Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answercited by 2
  18. 2025Focus on Likely Classes for Test-Time Predictioncited by 1
  19. 2025BD at BEA 2025 Shared Task: MPNet Ensembles for Pedagogical Mistake Identification and Localization in AI Tutor Responsescited by 1
  20. 2026Unlocking the Pre-Trained Model as a Dual-Alignment Calibrator for Post-Trained LLMscited by 1
  21. 2026The Well-Tempered Classifier: Some Elementary Properties of Temperature Scalingcited by 1
  22. 2026Ranking-Aware Calibration for Reliable Multimodal Reinforcement Learningcited by 1
  23. 2026Failed Reasoning Traces Tell You What Is Fixable (But Not by Reading Them)cited by 1
  24. 2026Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMscited by 1
  25. 2026Audited Conformal Prediction for Classification under Unknown Distribution Shiftcited by 1
  26. 2026Do Large Language Models Always Tell The Same Stories?cited by 1
  27. 2024Optical aberrations in autonomous driving: Physics-informed parameterized temperature scaling for neural network uncertainty calibrationcited by 0
  28. 2025MMD-Flagger: Leveraging Maximum Mean Discrepancy to Detect Hallucinationscited by 0
  29. 2025The Effects of Data Augmentation on Confidence Estimation for LLMscited by 0
  30. 2025Enabling Fine-Grained Operating Points for Black-Box LLMscited by 0
  31. 2026Efficient Self-Evaluation for Diffusion Language Models via Sequence Regenerationcited by 0
  32. 2026CHiL(L)Grader: Calibrated Human-in-the-Loop Short-Answer Gradingcited by 0
  33. 2026Entropy and Attention Dynamics in Small Language Models: A Trace-Level Structural Analysis on the TruthfulQA Benchmarkcited by 0
  34. 2026BiCon-Gate: Consistency-Gated De-colloquialisation for Dialogue Fact-Checkingcited by 0
  35. 2026Log-Likelihood, Simpson's Paradox, and the Detection of Machine-Generated Textcited by 0
  36. 2026Task-Aware Calibration: Provably Optimal Decoding in LLMscited by 0
  37. 2026Inducing Artificial Uncertainty in Language Modelscited by 0
  38. 2026ECUASn: A family of metrics for principled evaluation of uncertainty-augmented systemscited by 0

50 most cited of 58 All 58 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Scaling with Confidence: Calibrating Confidence of LLMs for Adaptive Test Time Scaling
  2. 2025Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach
  3. 2025Exploring the Impact of Temperature on Large Language Models:Hot or Cold?
  4. 2023On the Calibration of Large Language Models and Alignment
  5. 2024Calibrating Long-form Generations from Large Language Models
  6. 2026Calibrating Beyond English: Language Diversity for Better Quantized Multilingual LLM
  7. 2024Does Alignment Tuning Really Break LLMs' Internal Confidence?
  8. 2025On the Role of Temperature Sampling in Test-Time Scaling