mlsanityData to 7 Oct 2026

Non-Determinism of "Deterministic" LLM Settings

arXiv 2408.04667 ↗First posted 6 Aug 2024cs.CL cs.AI cs.LGcited by 84

Not on the map. Why?

Abstract (arXiv)

LLM (large language model) practitioners commonly notice that outputs can vary for the same inputs under settings expected to be deterministic. Yet the questions of how pervasive this is, and with what impact on results, have not to our knowledge been systematically investigated. We investigate non-determinism in five LLMs configured to be deterministic when applied to eight common tasks in across 10 runs, in both zero-shot and few-shot settings. We see accuracy variations up to 15% across naturally occurring runs with a gap of best possible performance to worst possible performance up to 70%. In fact, none of the LLMs consistently delivers repeatable accuracy across all tasks, much less identical output strings. Sharing preliminary results with insiders has revealed that non-determinism perhaps essential to the efficient use of compute resources via co-mingled data in input buffers so this issue is not going away anytime soon. To better quantify our observations, we introduce metrics focused on quantifying determinism, TARr@N for the total agreement rate at N runs over raw output, and TARa@N for total agreement rate of parsed-out answers. Our code and data are publicly available at https://github.com/breckbaldwin/llm-stability.

Key facts. Cited by 84 papers in the map (data through 7 Oct 2026). Builds on 19 of its 33 references that are in the map. First posted on arXiv 6 Aug 2024.

Builds on

19 of 33 references.

References · 19

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  4. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  5. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,767
  6. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  7. 2024A Survey on Large Language Models for Code Generationcited by 646
  8. 2022Reasoning with Language Model Prompting: A Surveycited by 251
  9. 2022Leveraging Large Language Models for Multiple Choice Question Answeringcited by 199
  10. 2024Are We Done with MMLU?cited by 178
  11. 2024Lessons from the Trenches on Reproducible Evaluation of Language Modelscited by 172
  12. 2023An Empirical Study of the Non-determinism of ChatGPT in Code Generationcited by 134
7 more
  1. 2023Quantifying Uncertainty in Answers from any Language Model and Enhancing their Trustworthinesscited by 121
  2. 2024The Good, The Bad, and The Greedy: Evaluation of LLMs Should Not Ignore Non-Determinismcited by 113
  3. 2023MultiLegalPile: A 689GB Multilingual Legal Corpuscited by 59
  4. 2023Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inferencecited by 40
  5. 2023The ART of LLM Refinement: Ask, Refine, and Trustcited by 29
  6. 2020Contextual Temperature for Language Modelingcited by 25
  7. 2023How Good Are LLMs at Out-of-Distribution Detection?cited by 15

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 1; 2025: 27; 2026: 56.

84 papers.

Sort:

By area:

  1. 2025A Sober Look at Progress in Language Model Reasoning: Pitfalls and Paths to Reproducibilitycited by 89
  2. 2025RobustFlow: Towards Robust Agentic Workflow Generationcited by 19
  3. 2025On the Effect of Sampling Diversity in Scaling LLM Inferencecited by 18
  4. 2025Evaluating Long-Context Reasoning in LLM-Based WebAgentscited by 15
  5. 2025Taming Uncertainty via Automation: Observing, Analyzing, and Optimizing Agentic AI Systemscited by 12
  6. 2026ACRFence: Preventing Semantic Rollback Attacks in Agent Checkpoint-Restorecited by 9
  7. 2025Challenges in Testing Large Language Model Based Software: A Faceted Taxonomycited by 8
  8. 2025Evaluation and Facilitation of Online Discussions in the LLM Era: A Surveycited by 8
  9. 2025ATLAS: Adaptive Trading with LLM AgentS Through Dynamic Prompt Optimization and Multi-Agent Coordinationcited by 7
  10. 2026Toward Expert Investment Teams:A Multi-Agent LLM System with Fine-Grained Trading Taskscited by 7
  11. 2025LLM Output Drift: Cross-Provider Validation & Mitigation for Financial Workflowscited by 6
  12. 2025STED and Consistency Scoring: A Framework for Evaluating LLM Structured Output Reliabilitycited by 5
38 more
  1. 2026Accuracy, Stability, and Repeated-Run Reliability of Large Language Models on Deterministic Programming Taskscited by 4
  2. 2025Through the Prism of Culture: Evaluating LLMs' Understanding of Indian Subcultures and Traditionscited by 3
  3. 2025The Non-Determinism of Small LLMs: Evidence of Low Answer Consistency in Repetition Trials of Standard Multiple-Choice Benchmarkscited by 3
  4. 2025Can LLMs Make (Personalized) Access Control Decisions?cited by 3
  5. 2026Beyond Reproducibility: Token Probabilities Expose Large Language Model Nondeterminismcited by 3
  6. 2026Multi-Agent Taint Specification Extraction for Vulnerability Detectioncited by 3
  7. 2026Compiled AI: Deterministic Code Generation for LLM-Based Workflow Automationcited by 3
  8. 2026Efficient Provably Secure Linguistic Steganography via Range Codingcited by 3
  9. 2026EQUITRIAGE: A Fairness Audit of Gender Bias in LLM-Based Emergency Department Triagecited by 3
  10. 2026FLIPS: Instance-Fingerprinting for LLMs via Pseudo-random Sequencescited by 3
  11. 2024Evaluating and Advancing Multimodal Large Language Models in Perception Ability Lenscited by 2
  12. 2025Designing Synthetic Discussion Generation Systems: A Case Study for Online Facilitationcited by 2
  13. 2025Something Just Like TRuST : Toxicity Recognition of Span and Targetcited by 2
  14. 2025LLM Based Bayesian Optimization for Prompt Searchcited by 2
  15. 2025ATA: A Neuro-Symbolic Approach to Implement Autonomous and Trustworthy Agentscited by 2
  16. 2025T-pro 2.0: An Efficient Russian Hybrid-Reasoning Model and Playgroundcited by 2
  17. 2026A Guide to Large Language Models in Modeling and Simulation: From Core Techniques to Critical Challengescited by 2
  18. 2026Cross-Context Verification: Hierarchical Detection of Benchmark Contamination through Session-Isolated Analysiscited by 2
  19. 2026Anchored Sliding Window: Toward Robust and Imperceptible Linguistic Steganographycited by 2
  20. 2026Introducing Background Temperature to Characterise Hidden Randomness in Large Language Modelscited by 2
  21. 2026Reasoning effort, not tool access, buys first-try reliability in agentic code generation: an observational studycited by 2
  22. 2026The Dice Roll Method: A Standardized Protocol for Repeated-Query Auditing of Large Language Model Brand Recommendationscited by 2
  23. 2025Event Segmentation Applications in Large Language Model Enabled Automated Recall Assessmentscited by 1
  24. 2025Reading Smiles: Proxy Bias in Foundation Models for Facial Emotion Recognitioncited by 1
  25. 2025MetaCipher: A Time-Persistent and Universal Multi-Agent Framework for Cipher-Based Jailbreak Attacks for LLMscited by 1
  26. 2025The Trust in AI-Generated Health Advice (TAIGHA) Scale and Short Version (TAIGHA-S): Development and Validation Studycited by 1
  27. 2026The Stability Trap: Evaluating the Reliability of LLM-Based Instruction Adherence Auditingcited by 1
  28. 2026Communication Enhances LLMs' Stability in Strategic Thinkingcited by 1
  29. 2026Beyond Relevance: On the Relationship Between Retrieval and RAG Information Coveragecited by 1
  30. 2026Like a Hammer, It Can Build, It Can Break: Large Language Model Uses, Perceptions, and Adoption in Cybersecurity Operations on Redditcited by 1
  31. 2026Iterating Toward Better Search: A Two-Agent Simulation Framework for Evaluating Agentic Search Architectures in E-Commercecited by 1
  32. 2026A Large-Scale Multi-Dimensional Empirical Study of LLMs for Conversation Summarizationcited by 1
  33. 2026A Reliability Assessment of LALM Audio Judges for Full-Duplex Voice Agentscited by 1
  34. 2026CAFE: A Compound-AI Factorial Evaluation Frameworkcited by 1
  35. 2026The Language of the Question Selects the Market: Query Language and Exit IP as Separable Factors in Commercial Recommendations from a Generative Search Interfacecited by 1
  36. 2025AutoTestForge: A Multidimensional Automated Testing Framework for Natural Language Processing Modelscited by 0
  37. 2025LAMP: Extracting Local Decision Surfaces From Large Language Modelscited by 0
  38. 2025BeLLMan: Controlling LLM Congestioncited by 0

50 most cited of 84 All 84 (JSON)