Blizado/discolm-kunoichi-7b-german-v0.1

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Jan 21, 2024Architecture:Transformer0.0K Warm

Blizado/discolm-kunoichi-7b-german-v0.1 is an experimental 7 billion parameter language model created by Blizado using the SLERP merge method. It combines SanjiWatsuki/Kunoichi-DPO-v2-7B and DiscoResearch/DiscoLM_German_7b_v1 to enhance German language quality and roleplay capabilities. This model is specifically optimized for generating grammatically sound German text, particularly excelling in German roleplay scenarios. Its 4096-token context length supports nuanced and extended interactions in German.

Loading preview...

Model Overview

Blizado/discolm-kunoichi-7b-german-v0.1 is a 7 billion parameter experimental language model developed by Blizado. It was created using the SLERP merge method to combine two distinct base models: SanjiWatsuki/Kunoichi-DPO-v2-7B and DiscoResearch/DiscoLM_German_7b_v1.

Key Capabilities

  • Enhanced German Language Quality: The merge specifically targets improving grammatical accuracy and natural-sounding German, building upon DiscoLM German 7B's strong performance in this area.
  • Optimized for German Roleplay: By integrating Kunoichi-DPO-v2-7B, which is trained for roleplay, the model aims to deliver superior performance in German roleplaying scenarios.
  • Reduced Grammatical Errors: Addresses common grammatical issues found in other German models, offering a more polished output.

Why this Merge?

The primary motivation behind this merge was to create a German model that combines the excellent grammatical proficiency of DiscoLM German 7B with the robust roleplay capabilities of Kunoichi DPO v2 7B. Initial testing has shown promising results in achieving a better overall German model, especially for interactive and roleplay-focused applications. The merge configuration utilized settings inspired by oshizo/japanese-e5-mistral-7b_slerp.