贪婪解码并非精度不变：大语言模型推理中的跨精度输出分歧

Computer Science > Machine Learning arXiv:2609.26621 (cs) [Submitted on 22 Sep 2026] Title:Greedy Decoding Is Not Precision-Invariant: Cross-Precision Output Divergence in LLM Inference Authors:Gaoyuan Du, Anam Nawaz Khan, Rex Zhou, Xiaoyang Liu, Deepayan Chakrabarti, Fnu Suya, Xueping Li View a PDF of the paper titled Greedy Decoding Is Not Precision-Invariant: Cross-Precision Output Divergence in LLM Inference, by Gaoyuan Du and 6 other authors View PDF Abstract:Greedy decoding from large language models is commonly treated as deterministic. We show it is not precision-invariant: the same model, prompt, and decoding algorithm produce different outputs in BF16 versus FP16 on identical hardware. Across our evaluations of six models (1.1B-7B parameters, four families; divergence additionally characterised at 12B) and three benchmarks, 49-100\% of prompts diverge; a single token flip often cascades into trajectory-level divergence. We develop an empirical error-propagation analysis and find that 22 layers of accumulated body error do not distinguish flipping from non-flipping steps; the outcome depends primarily on the top-two logit margin at the LM head relative to the directional perturbation between the top-two candidates. The analysis makes five testable predictions about intervention outcomes, including that applying more FP32 compute (broader scope) makes agreement worse. The experiments match all five predictions. The best-performing low-overhead intervention we evaluate, selective FP32 LM head recomputation, triggered only when the margin falls below a threshold, delivers +22-36 pp exact agreement on A10G (+12-21 pp on L4 and A100) at less than 4\% latency overhead in low-batch (batch size =8 and under end-to-end FP8 in our tests. Comments: Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2609.26621 [cs.LG] (or arXiv:2609.26621v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2609.26621 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Journal reference: Transactions on Machine Learning Research, 2026 Submission history From: Gaoyuan Du [view email] [v1] Tue, 22 Sep 2026 15:54:16 UTC (1,455 KB) Full-text links: Access Paper: View a PDF of the paper titled Greedy Decoding Is Not Precision-Invariant: Cross-Precision Output Divergence in LLM Inference, by Gaoyuan Du and 6 other authors View PDF TeX Source view license Current browse context: cs.LG new | recent | 2026-09 Change to browse by: cs cs.AI References & Citations NASA ADS Google Scholar Semantic Scholar export BibTeX citation Loading... BibTeX formatted citation × loading... Data provided by: Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?) IArxiv recommender toggle IArxiv Recommender (What is IArxiv?) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs. Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)