Hugging Face Overview
QVikhr-3-4B-Instruction: Bilingual Performance for Russian and English
QVikhr-3-4B-Instruction is a 4 billion parameter instruction-tuned model built upon the robust Qwen3-4B architecture. Developed by Vikhrmodels, its primary distinction lies in its specialized training on the GrandMaster2 dataset, a large Russian-language corpus, using Supervised Fine-Tuning (SFT).
Key Capabilities & Features
- Bilingual Proficiency: Optimized for high-efficiency text processing in both Russian (RU) and English (EN).
- Instruction Following: Designed to generate precise, context-sensitive responses and execute tasks based on instructions.
- Enhanced Russian Performance: Achieves a score of 78.2 on the Ru Arena General benchmark, significantly outperforming its base model, Qwen3-4B (64.8).
- Quantized Variants: Available in GGUF and MLX (4-bit and 8-bit) formats for flexible deployment.
Ideal Use Cases
- Russian Language Applications: Excellent for professional environments requiring accurate text analysis and generation in Russian.
- Bilingual Systems: Suitable for applications needing seamless instruction processing and contextual responses in both Russian and English.
- Instructional Learning Tasks: Optimized for scenarios where models need to follow complex instructions effectively.