🤖 AI Summary
In large-scale recommendation systems, conventional hash indexing suffers from embedding collisions and stale embedding inheritance due to high-cardinality IDs, degrading model performance and personalization quality. This work proposes MPZCH, a novel mechanism that integrates multi-probe linear probing, configurable probing strategies, and an active eviction policy, leveraging auxiliary tensors and efficient CUDA kernels to nearly eliminate embedding collisions within a practical table size. MPZCH resets and recycles evicted slots, ensuring new features are learned from scratch. To the best of our knowledge, this is the first approach to achieve near-zero-collision embedding indexing in production-scale systems, substantially improving embedding freshness and quality while maintaining training throughput (QPS) and inference latency comparable to existing methods. The implementation has been integrated into the open-source TorchRec library.
📝 Abstract
Embedding tables are critical components of large-scale recommendation systems, facilitating the efficient mapping of high-cardinality categorical features into dense vector representations. However, as the volume of unique IDs expands, traditional hash-based indexing methods suffer from collisions that degrade model performance and personalization quality. We present Multi-Probe Zero Collision Hash (MPZCH), a novel indexing mechanism based on linear probing that effectively mitigates embedding collisions. With reasonable table sizing, it often eliminates these collisions entirely while maintaining production-scale efficiency. MPZCH utilizes auxiliary tensors and high-performance CUDA kernels to implement configurable probing and active eviction policies. By retiring obsolete IDs and resetting reassigned slots, MPZCH prevents the stale embedding inheritance typical of hash-based methods, ensuring new features learn effectively from scratch. Despite its collision-mitigation overhead, the system maintains training QPS and inference latency comparable to existing methods. Rigorous online experiments demonstrate that MPZCH achieves zero collisions for user embeddings and significantly improves item embedding freshness and quality. The solution has been released within the open-source TorchRec library for the broader community.