IlyaGusev/saiga_gemma3_12b

Warm
Public
License: gemma
Hugging Face
Overview

Overview

IlyaGusev/saiga_gemma3_12b is a 12 billion parameter language model, fine-tuned by IlyaGusev from the mlabonne/gemma-3-12b-it-abliterated base model. It is specifically designed for Russian language processing and functions as an automatic assistant. The model utilizes a custom prompt format based on Gemma3, incorporating system messages for structured conversations.

Key Capabilities

  • Russian Language Proficiency: Optimized for generating and understanding Russian text.
  • Conversational AI: Fine-tuned to act as an interactive assistant, capable of engaging in dialogues.
  • Extended Context Window: Supports a context length of 32768 tokens, allowing for more extensive conversations and complex queries.
  • Performance: Evaluation results show competitive performance on Russian benchmarks like PingPong and RuArenaHard, with comparisons against models such as gpt-4o.

Good For

  • Developing Russian-speaking chatbots and virtual assistants.
  • Applications requiring detailed Russian text generation and comprehension.
  • Research and development in Russian natural language processing, particularly for conversational tasks.