Vikhr-Qwen-2.5-1.5B-Instruct Overview

Vikhr-Qwen-2.5-1.5B-Instruct is a 1.5 billion parameter instruction-tuned language model from Vikhrmodels, built upon the Qwen-2.5-1.5B-Instruct architecture. Its primary distinction is its bilingual specialization in Russian and English, achieved through supervised fine-tuning (SFT) on the proprietary Russian-language GrandMaster-PRO-MAX dataset.

Key Capabilities

• Bilingual Proficiency: Optimized for high-efficiency text processing and understanding in both Russian and English.
• Instruction Following: Designed to generate precise responses based on given instructions.
• Contextual Responses: Capable of providing coherent and contextually relevant answers.
• Text Analysis: Excels in analyzing and interpreting textual data.
• Training Methodology: Fine-tuned using a synthetic dataset of 150k instructions, incorporating Chain-Of-Thought (CoT) and GPT-4-turbo prompts to enhance accuracy and coherence.

Good For

• Applications requiring robust performance in Russian language tasks.
• Use cases demanding accurate instruction generation and contextual understanding in a bilingual (RU/EN) setting.
• Integration into user-facing applications and services where efficient text processing is crucial.

Quantized variants (GGUF, MLX 4-bit, MLX 8-bit) are also available for optimized deployment.