From Data to Device: ELMOD An Efficient German-First 2.7B Language Model for Mobile Inference
Source: https://arxiv.org/abs/2607.24585
arXiv: 2607.24585
Authors: Darina Gold, Alexander Schwirjow, Viktor Haag, Viktor Hangya, Joel Schlotthauer, Fabian Küch, Luzian Hahn
Published: 2026-07-27
Categories: cs.CL
PDF: https://arxiv.org/pdf/2607.24585v1
Abstract (en)
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.
Summary (zh)
ELMOD 是面向资源受限/移动推理的 2.7B 德文优先模型，训练预算约 55k H100 GPU hours，仅用公开数据。相对英文导向流水线，作者强调德语形态变化、复合词与拼写习惯的预处理，并加入质量过滤与改写以提升指令质量、改善 annealing、降低总算力。摘要称其在 <3B 量级最强，德语任务可匹配 7B 级表现。对手机本地助手与中文端侧小模型数据路径有对照价值。
One-liner
端侧 2.7B 仍有空间：ELMOD 用语言特化预处理与公有数据，在 <3B 德文任务摸到 7B 级表现。
Obsidian evidence
OpenClaw定时任务/论文流水线/2026-07-29-论文流水线.md- opencli:
arxiv paper 2607.24585 -f json - Run date: 2026-07-29