Newest papers

How Developers Configure LLM Inference: A Large-Scale Study of Real-World Practices

Nada Zine, Matías Martínez, Romain Rouvoy, Clément Quinton · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.27945 · cited by 0 works (OpenAlex) · open-access PDF

Russian

LLMs все чаще интегрируются в программные системы, делая развертывание и конфигурацию выводного контекста рецидивирующей задачей программного обеспечения. Эти резервные коды выявляют много вариантов, которые влияют на загрузку моделей, генерацию, обслуживание и обработку запросов. Тем не менее мало известно о том, какие из этих резервных кодов разработчики эксплицитно конфигурируют на практике. В этой статье мы представляем масштабное эмпирическое исследование практик конфигурации вывода LLM в программном обеспечении с открытым исходным кодом, используя Hugging Face Transformers, vLLM и OpenAI API. Мы анализируем 5,398 резервных кодов GitHub, используя по крайней мере один из этих резервных кодов, и определяем 3

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

Abstract

LLMs are increasingly integrated into software systems, making deployment and inference configuration a recurring software engineering task. Inference backends expose many options that affect model loading, generation, serving, and request processing. Yet little is known about which of these options developers explicitly configure in practice. In this paper, we present a large-scale empirical study of LLM inference configuration practices in open-source software using Hugging Face Transformers, vLLM, and the OpenAI API. We analyze 5,398 GitHub repositories using at least one of these backends and identify 345,117 explicitly configured option occurrences. Our results show that developers use only a small part of the available configuration space regularly. Although 98.8% of the documented backend-specific options appear at least once, 78.9% are configured in at most 10% of repositories using the corresponding backend. Configurations are also small, with a median of 2 options for Transformers and 4 for vLLM and OpenAI, but the combinations of options vary widely: 56.8-61.6% of multi-option signatures appear in only one repository. We also observe recurring option values as well as substantial variation in how values are chosen. Finally, configuration practices differ across application contexts, with option-adoption differences reaching 30.7 percentage points. These findings characterize the configuration surface developers make explicit rather than the importance or optimality of individual options. They provide empirical evidence for prioritizing future studies of inference configuration and for improving backend documentation and configuration support.

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

Topics: Software deployment, Documentation, Inference, Computer science, Software engineering, Software

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