IAG-Transformer is a graph Transformer architecture specifically designed for part machining feature recognition. Its core innovation lies in introducing instance-aware attention bias into the self-attention mechanism, enabling the model to explicitly model the semantic relationships between faces within the same machining feature instance.