Knowledge Graph Completion aims to infer missing triples and other information in knowledge graphs. However, existing methods often incur high computational costs and are typically restricted to addressing only a single task. To address these issues, we propose a Transformer-based framework called TTKGC(Transformers for Three Knowledge Graph Completion tasks), which is capable of simultaneously handling three knowledge graph completion tasks: entity typing, relation prediction, and link prediction. We comprehensively analyze the logical relationships and inner connections among these tasks and propose a progressive training method that allows downstream tasks to benefit from the model parameters of upstream tasks. Additionally, we introduce an adaptive optimization strategy to improve the model’s training efficiency. The results of quantitative analysis and ablation experiments demonstrate that the proposed framework and strategies are effective, with the model achieving competitive performance on benchmark datasets.