# Russian instruction optimization

QVikhr-3-1.7B-Instruction-noreasoning is an instruction model based on Qwen/Qwen3-1.7B, trained on the Russian dataset GrandMaster2, and is specifically designed for efficient processing of Russian and English texts.

This is an instruction - type version that adapts the Qwen2.5 - 32B model to Russian. By replacing the tokenizer and applying LEP technology, the generation speed of Russian text is significantly improved.

T-lite-instruct-0.1 is an instruction fine-tuned version based on the T-lite-0.1 model, specifically designed for Russian tasks and requires further fine-tuning before use

This is an instruction-following model specifically optimized for Russian, based on the Vikhr Base architecture, trained on various instruction datasets using curriculum learning.