Versioned Late Materialization for Ultra-Long Sequence Training in Recommendation Systems at Scale
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917442235662336 |
|---|---|
| author | Guo, Liang Song, Ge Deng, Litao Sun, Jianhui Hu, Chufeng Zhang, Lu Ma, Zhen Chen, Shouwei Liu, Weiran Sreeshylan, Sarang Masti Meng, Xiaoxuan |
| author_facet | Guo, Liang Song, Ge Deng, Litao Sun, Jianhui Hu, Chufeng Zhang, Lu Ma, Zhen Chen, Shouwei Liu, Weiran Sreeshylan, Sarang Masti Meng, Xiaoxuan |
| contents | Modern Deep Learning Recommendation Models (DLRMs) follow scaling laws with sequence length, driving the frontier toward ultra-long User Interaction History (UIH). However, the industry-standard "Fat Row" paradigm, which pre-materializes these sequences into every training example, creates a storage and I/O wall where data infrastructure usage exceeds GPU training capacity due to data redundancy that is amplified in multi-tenant environments where models with vastly different sequence length requirements share a union dataset. We present a \emph{versioned late materialization} paradigm that eliminates this redundancy by storing UIH once in a normalized, immutable tier and reconstructing sequences just-in-time during training via lightweight versioned pointers. The system ensures Online-to-Offline (O2O) consistency through a bifurcated protocol that prevents future leakage across both streaming and batch training, while a read-optimized immutable storage layer provides multi-dimensional projection pushdown for heterogeneous model tenants. Disaggregated data preprocessing with pipelined I/O prefetching and data-affinity optimizations masks the latency of training-time sequence reconstruction, keeping training throughput compute-bound by GPUs. Deployed on production DLRMs, the system reduces training data infrastructure resource usage while enabling aggressive sequence length scaling that delivers significant model quality gains, serving as the foundational data infrastructure for modern recommendation model architectures, including HSTU and ULTRA-HSTU. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_24806 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Versioned Late Materialization for Ultra-Long Sequence Training in Recommendation Systems at Scale Guo, Liang Song, Ge Deng, Litao Sun, Jianhui Hu, Chufeng Zhang, Lu Ma, Zhen Chen, Shouwei Liu, Weiran Sreeshylan, Sarang Masti Meng, Xiaoxuan Information Retrieval Artificial Intelligence Databases Modern Deep Learning Recommendation Models (DLRMs) follow scaling laws with sequence length, driving the frontier toward ultra-long User Interaction History (UIH). However, the industry-standard "Fat Row" paradigm, which pre-materializes these sequences into every training example, creates a storage and I/O wall where data infrastructure usage exceeds GPU training capacity due to data redundancy that is amplified in multi-tenant environments where models with vastly different sequence length requirements share a union dataset. We present a \emph{versioned late materialization} paradigm that eliminates this redundancy by storing UIH once in a normalized, immutable tier and reconstructing sequences just-in-time during training via lightweight versioned pointers. The system ensures Online-to-Offline (O2O) consistency through a bifurcated protocol that prevents future leakage across both streaming and batch training, while a read-optimized immutable storage layer provides multi-dimensional projection pushdown for heterogeneous model tenants. Disaggregated data preprocessing with pipelined I/O prefetching and data-affinity optimizations masks the latency of training-time sequence reconstruction, keeping training throughput compute-bound by GPUs. Deployed on production DLRMs, the system reduces training data infrastructure resource usage while enabling aggressive sequence length scaling that delivers significant model quality gains, serving as the foundational data infrastructure for modern recommendation model architectures, including HSTU and ULTRA-HSTU. |
| title | Versioned Late Materialization for Ultra-Long Sequence Training in Recommendation Systems at Scale |
| topic | Information Retrieval Artificial Intelligence Databases |
| url | https://arxiv.org/abs/2604.24806 |