Kolibri Vikhr Mistral 0427

🚀 Instruction Following Model for Russian Language

This is an instruction following model optimized for the Russian language, based on Vikhr Base. It offers capabilities in information retrieval, classification, semantic analysis, and English - Russian translation. Currently, it's in the pre - alpha stage.

🚀 Quick Start

This model is trained using kolibrify through curriculum learning on multiple instruction datasets. It uses the ChatML template and is designed to be sensitive to system prompts. You can experiment with different system prompts to get better results.

✨ Features

• Multifunctional: Suitable for various tasks such as information retrieval, classification, semantic analysis, and English - Russian translation.
• Curriculum Learning: Trained via curriculum learning on a wide range of instruction datasets.
• ChatML Template: Utilizes the ChatML template and is sensitive to system prompts.

📦 Installation

No installation steps were provided in the original README, so this section is skipped.

💻 Usage Examples

No code examples were provided in the original README, so this section is skipped.

📚 Documentation

Instruction following evals

The model was tested using the following benchmarks:

• ruIFEval
• ifeval

🔧 Technical Details

No specific technical details (more than 50 words) were provided in the original README, so this section is skipped.

📄 License

This project is licensed under the Apache - 2.0 license.