Through pre-training and fine-tuning, our AI team has constructed a large model specialized in state-owned enterprise knowledge. From 2024 to the first half of 2025, we conducted two rounds of vertical domain training. Evaluation metrics have surpassed those of the base model and meet user requirements.