Skarmorie/Mag-Mell-RU-035: Enhanced Russian Language Model

This model, Skarmorie/Mag-Mell-RU-035, is a 12 billion parameter language model created through a sophisticated merge process. It leverages the Linear DARE merge method, building upon the robust foundation of inflatebot/MN-12B-Mag-Mell-R1.

Key Capabilities

• Improved Russian Output: The primary differentiator of this model is its enhanced stability and quality for Russian language generation. This improvement stems from the integration of Aleteian/base-ground-2, which itself is a merge of Saiga and Vikhr models, known for their Russian language proficiency.
• Merge Method: Utilizes the Linear DARE method, a technique designed for effective merging of pre-trained language models.
• Base Model: inflatebot/MN-12B-Mag-Mell-R1 serves as the foundational model, providing a strong general language understanding.

Good For

• Applications requiring stable and high-quality text generation in Russian.
• Developers looking for a 12B parameter model with a specific focus on multilingual performance, particularly for Russian.
• Use cases where the combined strengths of Mag-Mell and Russian-optimized models are beneficial.