Vikhrmodels/Vikhr-Gemma-2B-instruct

Warm
Public
2.6B
BF16
8192
Aug 20, 2024
License: apache-2.0
Hugging Face
Overview

Vikhr-Gemma-2B-instruct: Russian Language Specialization

Vikhr-Gemma-2B-instruct is a 2.6 billion parameter instruction-tuned language model built upon the Gemma 2 2B base architecture. Developed by Vikhrmodels, its primary differentiator is its specialized training on the extensive Russian-language dataset, GrandMaster-PRO-MAX. This focused training enables the model to process and generate Russian text with high proficiency.

Key Capabilities

  • Russian Language Expertise: Specifically fine-tuned for the Russian language, making it highly effective for tasks in this domain.
  • Compact and Powerful: Offers strong performance for its 2.6 billion parameter size, making it efficient for deployment.
  • Instruction Following: Designed to understand and execute instructions, providing coherent and relevant responses.

Performance Insights

On the ru_arena_general benchmark, Vikhr-Gemma-2B-instruct achieved a score of 45.82, demonstrating competitive performance among other models, particularly within the Russian language context. Its base model is google/gemma-2-2b-it.

Good For

  • Applications requiring robust Russian language generation and understanding.
  • Instruction-following tasks in Russian.
  • Developers looking for a compact yet powerful model specialized for the Russian linguistic landscape.