From Data to Device: ELMOD An Efficient German-First 2.7B Language Model for Mobile Inference
2026-07-27 • Computation and Language
Computation and Language
AI summaryⓘ
The authors created ELMOD, a small German language model with 2.7 billion parameters, designed to run efficiently on devices with limited computing power. They trained it using only publicly available data and special methods tailored to German language features like word formation and spelling. They also improved the training data by filtering and rephrasing it to make the instructions clearer and easier to learn from. Their approach allowed ELMOD to perform as well as much larger models with over twice the number of parameters.
language modelGerman languageon-device deploymentdata preprocessingmorphological variationcompoundingorthographic conventionsquality filteringmodel sizeinference efficiency
Authors
Darina Gold, Alexander Schwirjow, Viktor Haag, Viktor Hangya, Joel Schlotthauer, Fabian Küch, Luzian Hahn
Abstract
We present ELMOD - Efficient Language Model for On-Device Deployment - a compact (2.7B) German language model designed for efficient inference on resource-constrained hardware. ELMOD was trained on a limited computational budget (55k H100 GPU hours) using exclusively publicly available data. We developed a suite of German-specific data pre-processing, which differ from English-oriented counterparts in their handling of morphological variation, compounding, and orthographic conventions. Furthermore, we introduced a quality filtering and rephrasing step, which increased the instructional quality of the data, improved performance during the annealing phase, and reduced overall compute requirements. Thanks to our architectural model and data choices, including prefiltering, our educational-quality filtering and rephrasal to raise the educational-quality, ELMOD is the strongest performer in its size class (<3B), matching the performance of 7B-parameter models in German.