Newest papers

Auditing Conformal Prediction for Language Models Under Inference-Time Model Shift

Achref Doula, Max Mühlhäuser · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28937 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Previsão de Conformidade (CP) para modelos de linguagem é tipicamente avaliada sob um protocolo correspondente, em que a calibração e a implantação usam a mesma configuração de inferência pós-treinamento, como o modo de quantificação e a temperatura de inferência. Na implantação, essas escolhas muitas vezes diferem daquelas usadas na calibração para atender às restrições de memória ou de transmissão ou políticas de serviço. Argumentamos que isso expõe uma limitação nos protocolos de avaliação de CP atuais para modelos de linguagem. Formalizamos o fenômeno subjacente como mudança de modelo de tempo de inferência: uma mudança na distribuição de pontuação de conformidade induzida pela configuração de inferência em vez da distribuição de input-label, e teoricamente caracterizamos como

Machine translation (Meta M2M100); the English original follows.

Abstract

Conformal prediction (CP) for language models is typically evaluated under a matched protocol, in which calibration and deployment use the same post-training inference configuration, such as quantization mode and inference temperature. In deployment, these choices often differ from those used at calibration to meet memory or throughput constraints or serving policies. We argue that this exposes a limitation in current CP evaluation protocols for language models. We formalize the underlying phenomenon as inference-time model shift: a change in the conformal score distribution induced by inference configuration rather than by the input--label distribution, and theoretically characterize how it breaks split-conformal exchangeability even when the task distribution and trained model are unchanged. We then evaluate its consequences across 8 language models, 4 finite-label benchmarks, and 12 inference configurations. We find that matched evaluation can substantially overstate deployment-time coverage, and that the effect is heterogeneous across models and datasets. We argue that inference configuration should be reported as a first-class experimental axis in CP evaluations for language-model systems.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Inference, Computer science, Fiducial inference, Artificial intelligence, Software deployment, Language model

Is this paper real? This page was built from OpenAlex and arXiv records, not generated. Software can verify any reference in one call: GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.28937, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txt

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange