VAGOsolutions/SauerkrautLM-SOLAR-Instruct

SauerkrautLM-SOLAR-Instruct Overview

SauerkrautLM-SOLAR-Instruct is a 10.7 billion parameter instruction-tuned model developed by VAGO solutions, building upon the Upstage SOLAR-10.7B-Instruct-v1.0 base. Its primary differentiator is its enhanced German language capabilities, achieved through fine-tuning with a specialized mix of German data augmentation and translated datasets. This process, including alignment via DPO with the German SauerkrautLM-DPO dataset, addresses the common issue of unnatural German phrasings often resulting from simple translation.

Key Capabilities

• Improved German Language Proficiency: Specifically trained to produce grammatically and syntactically correct German with natural wording.
• DPO Alignment: Utilizes Direct Preference Optimization with a unique German DPO dataset for refined instruction following.
• Multilingual Support: Supports both English and German, with a focus on German quality.
• Contamination-Free Training: Rigorous data contamination tests confirm the integrity of its training datasets, particularly for ARC, MMLU, TruthfulQA, and GSM8K.

Good For

• Applications requiring high-quality German text generation and understanding.
• Use cases where a robust, instruction-following model with strong German linguistic accuracy is crucial.
• Developers seeking a 10.7B parameter model that balances general performance with specialized German language optimization.