t-tech/T-pro-it-1.0

T-pro-it-1.0 is a 32.8 billion parameter model developed by t-tech, built upon the Qwen 2.5 model family with a 131072 token context length. It incorporates continual pre-training and alignment techniques, specifically optimized for Russian language tasks. The model demonstrates strong performance across various Russian benchmarks, particularly excelling in mathematical reasoning (ruGSM8K, ruMATH) and coding tasks (ruCodeEval) compared to other open-source models.