Industrial short video ranking improves with chain dimension scaling
Chain Dimension Scaling: Communication-Efficient Loop-Style Scaling for Industrial Short-Video Ranking
Artificial Intelligence
Summary
Recommender systems for short videos have grown very large, making them slow and expensive to train. The authors propose a new way to scale these models by repeatedly using the same part many times, which doesn’t add more parameters but helps the model get better. They show this approach can either reduce the size and speed up training without losing accuracy or improve quality with the same number of parameters. This method also reduces communication costs between devices during training, making it more efficient in real-world systems.
What this means in practice
- •For machine learning engineers: Build more efficient large-scale ranking models that reduce training costs while maintaining or improving prediction quality using iterative parameter sharing.
- •For distributed system architects: Design training systems with lower cross-device communication overhead by employing iterative computation to scale models without increasing parameter counts.
Authors
Zhimeng Jiang, Yuening Li, Ruixiao Sun, Yuanzhen Lin, Kun Jin, Chuan He, Liang Liu
Abstract
Industrial recommendation rankers keep scaling in width and depth to billions of parameters, but this increasingly incurs costs in parameter growth, distributed communication overhead, and training inefficiency. We introduce \underline{C}hain \underline{D}imension \underline{S}caling (CDS), a scaling strategy orthogonal to width and depth. CDS repeatedly applies a single feature-interaction block with parameters shared across iterations, so the number of iterations (the chain length) becomes a new scaling dimension that adds no parameters. A lightweight intra-chain residual connection stabilizes optimization and improves quality. CDS enables two regimes. In the \emph{compression regime}, shallow blocks with a longer chain match deeper baselines with fewer parameters and faster training; in the \emph{quality-scaling regime}, increasing chain length under a fixed parameter budget yields consistent offline and online gains, suggesting the chain dimension adds modeling capacity beyond parameter reuse. We evaluate CDS in a large-scale multi-task short-video ranking system in production. In the compression regime, CDS stays within $0.03\%$ of baseline offline quality while reducing trainable parameters by 26.6\% and improving training throughput by 6.4\%, with neutral-to-positive online participation. In the quality-scaling regime, quality improves monotonically with chain length. Wall-clock analysis shows the efficiency gain comes mainly not from reduced FLOPs but from lower cross-device synchronization overhead and reduced optimizer update cost. These results position parameter-shared iterative computation as a practical, communication-efficient scaling axis for industrial ranking systems.