![[2602.17697] Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17697] Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters
Summary
This article introduces a novel approach to optimizing inference hyperparameters in Large Language Models (LLMs) using variability modeling, addressing energy efficiency and sustainability concerns.
Why It Matters
As LLMs become integral to various applications, optimizing their inference processes is crucial for reducing energy consumption and enhancing sustainability. This research offers a systematic method to navigate the complex configuration space of LLMs, potentially leading to more efficient AI deployments.
Key Takeaways
- Variability modeling can simplify the analysis of LLM inference configurations.
- The study demonstrates how to evaluate hyperparameters' effects on energy consumption and performance.
- A new research direction is proposed that merges software engineering with machine learning.
Computer Science > Machine Learning arXiv:2602.17697 (cs) [Submitted on 6 Feb 2026] Title:Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters Authors:Nada Zine, Clément Quinton, Romain Rouvoy View a PDF of the paper titled Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters, by Nada Zine and 2 other authors View PDF HTML (experimental) Abstract:Large Language Models (LLMs) are being increasingly used across a wide range of tasks. However, their substantial computational demands raise concerns about the energy efficiency and sustainability of both training and inference. Inference, in particular, dominates total compute usage, making its optimization crucial. Recent research has explored optimization techniques and analyzed how configuration choices influence energy consumption. Yet, the vast configuration space of inference servers makes exhaustive empirical evaluation infeasible due to combinatorial explosion. In this paper, we introduce a new perspective on this problem by treating LLMs as configurable systems and applying variability management techniques to systematically analyze inference-time configuration choices. We evaluate our approach on the Hugging Face Transformers library by representing generation hyperparameters and their constraints using a feature-based variability model, sampling representative configurations, measuring their energy consumption, latency, accuracy, and learning predictive models from t...