Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters
Nada Zine, Clément Quinton, Romain Rouvoy · 2026
Read Pimp My LLM: Leveraging Variability Modeling to Tune Inference Hyperparameters by Nada Zine, Clément Quinton, Romain Rouvoy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17697.txt.
Description
Nada Zine and team treat large language models as tunable systems to tame run-time costs. Run time dominates total compute, so its tuning matters for energy and lasting use, yet the setup space of inference servers defies full testing. Variability models from software building offer an orderly alternative.
The team models output settings and their constraints with a feature-based variability model, using the Hugging Face library. They sample typical setups and measure energy use, latency and scores, then learn forecast models from the data. Results show variability models manage run-time tangle well, revealing setting effects, links and trade-offs. They also predict run-time conduct from limited measures, opening a bridge between software building and machine learning for green setup choices.
Catalog record
| CREATOR | Nada Zine, Clément Quinton, Romain Rouvoy |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17697