Saiga/Gemma2 9B: Russian Gemma-2 Chatbot
IlyaGusev/saiga_gemma2_9b is an instruction-tuned large language model developed by IlyaGusev, built upon the Google Gemma-2 9B Instruct architecture. This 9 billion parameter model is specifically designed and fine-tuned to function as a Russian-language automatic assistant, engaging in conversations and providing helpful responses.
Key Capabilities
- Russian Language Proficiency: Optimized for understanding and generating natural, fluent Russian text.
- Chatbot Functionality: Designed to handle conversational turns, acting as an interactive assistant.
- Gemma-2 Base: Leverages the capabilities of the Gemma-2 9B Instruct model for its foundational language understanding.
Training and Versions
The model has undergone several iterations, with v2 being the latest. Training involved Supervised Fine-Tuning (SFT) and KTO (Kahneman-Tversky Optimization) using specific dataset and model configurations, detailed in the README. Evaluation was conducted using the alpaca_eval framework with alpaca_eval_cot_gpt4_turbo_fn as the evaluator, comparing its performance against the base gemma_2_9b_it_abliterated model.
Good for
- Applications requiring a robust Russian-speaking AI assistant.
- Chatbot implementations for customer support or interactive services in Russian.
- Developers looking for a Gemma-2 based model with strong Russian language capabilities.