This German open-source LLM can out-reason GPT-3.5, proving sauerkraut might be brain food

DiscoLM German 7b is an open-source German language model based on Mistral. The model is optimized for German texts and understands, generates, and interacts effectively with German-language content, but also retains its English language capabilities and is supposed to be particularly good at translation tasks. In MT Bench, DiscoLM German 7b lags behind GPT-3.5 in writing, but according to the developer, the real quality of the model lies in its German expression, which is not covered by benchmarks, but felt by native speakers. DiscoLM is even ahead of GPT-3.5 in reasoning. The model is also trained for RAG (Retrieval Augmented Generation) applications.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top