Back to glossary

Term

SauerkrautLM

SauerkrautLM is a German-language LLM family from the German startup VAGOsolutions — fine-tunes based on Llama, Qwen, Mistral, and other open architectures.

SauerkrautLM — explained in more detail

SauerkrautLM is not a separate model architecture but a family of German-optimized fine-tunes built by the German startup VAGOsolutions on top of existing open models. Training combines German data augmentation with translated datasets, often paired with DPO (Direct Preference Optimization). The goal: grammatically and idiomatically clean German without sacrificing the underlying model’s capabilities.

Variants

  • SauerkrautLM-v1 (7B, 70B): first generation, standalone line.
  • SauerkrautLM-v2-14b-DPO: DPO-tuned variant with improved English while preserving German strength.
  • Llama-3-SauerkrautLM-8b/70b-Instruct: fine-tunes on Meta Llama 3.
  • SauerkrautLM-Qwen-32b: Qwen 2.5-based, collaboration with Hyperspace.ai.
  • SauerkrautLM-Nemo-12b-Instruct: Mistral-Nemo-based with Spectrum Fine-Tuning.
  • SauerkrautLM-1.5b: Qwen2-1.5B with Continuous Pre-Training for German.

Context

SauerkrautLM is relevant for applications where German is not just supported but the primary language — government, legal, or customer-dialogue scenarios with data protection requirements that demand local execution. The models are freely available on Hugging Face and can run locally via Ollama, LM Studio, or llama.cpp.

Entdecke mehr

Themenuebersicht