<p>Traffic forecasting is a key aspect in modern transportation systems, the majority of existing traffic forecasting models often struggle to concurrently address complex spatial dependencies and extended temporal dynamics. This study proposes STGT (Spatio-Temporal Graph Transformer), a traffic forecasting model that integrates graph attention mechanisms with causal temporal self-attention. Such an innovative integration efficiently address the limitations of conventional GCN-RNN hybrids and pure transformer-based models. Unlike to various existing models that depend on recurrent architectures or ignores the fundamental road network topology, STGT implements an adjacency-aware graph attention mechanism that explicitly acknowledges spatial connectivity, while its causal self-attention module effectively captures temporal dependencies. We evaluate STGT using two popular traffic datasets and show its improved performance relative to several existing models. The evaluation shows that the proposed STGT outperforms the existing models.</p>