RefalMachine/RuadaptQwen3-32B-Instruct is a 32 billion parameter instruction-tuned language model based on Qwen/Qwen3-32B, specifically adapted for the Russian language. It features a replaced tokenizer and continued pre-training on a Russian corpus, utilizing Learned Embedding Propagation (LEP) to significantly increase Russian text generation speed by up to 100%. This model is designed for high-speed and efficient processing of Russian-language tasks, functioning as a hybrid reasoner with an optional thinking mode.
No reviews yet. Be the first to review!