The pre-ranking stage of industrial recommender systems faces a fundamental conflict between efficiency and effectiveness. While powerful models like Target Attention (TA) excel at capturing complex feature interactions in the ranking stage, their high computational cost renders them unsuitable for pre-ranking, which relies on simple vector product models. This discrepancy creates a performance bottleneck for the overall system. To bridge this gap, this paper proposes TARQ, a novel pre-ranking framework. Inspired by generative models, TARQ's core innovation is to apply an architecture approximating TA to pre-ranking through residual quantization. This allows us to apply TA's modeling performance to the latency-critical pre-ranking stage for the first time, establishing a new state-of-the-art trade-off between accuracy and efficiency. Extensive offline experiments and large-scale online A/B testing on Taobao demonstrate TARQ's significant improvements in ranking performance. Consequently, our model has been fully deployed in production, serving tens of millions of daily active users and resulting in significant business improvements.
