LLäMmlein: Transparent, Compact and Competitive German-Only Language Models from Scratch
This work provides incremental improvements for the German NLP research community by offering open-source models and training data.
The authors tackled the problem of creating transparent and competitive German-only language models from scratch, resulting in two models (120M and 1B parameters) that performed competitively on benchmarks, matching or surpassing similar-sized models.
We create two German-only decoder models, LLäMmlein 120M and 1B, transparently from scratch and publish them, along with the training data, for the German NLP research community to use. The model training involved several key steps, including extensive data preprocessing, the creation of a custom German tokenizer, the training itself, as well as the evaluation of the final models on various benchmarks. Throughout the training process, multiple checkpoints were saved and analyzed using the SuperGLEBer benchmark to monitor the models' learning dynamics. Compared to state-of-the-art models on the SuperGLEBer benchmark, both LLäMmlein models performed competitively, consistently matching or surpassing models with similar parameter sizes. The results show that the models' quality scales with size as expected, but performance improvements on some tasks plateaued early, offering valuable insights into resource allocation for future model development.