Contents

• Introduction
  • Recommendation models
  • HybridBackend
• Architecture
• Data Loading
  • 1. Data Frame
  • 2. Tabular Dataset
  • 3. Deduplication
  • 4. Tips
  • 5. Benchmark
• Distributed Training
  • 1. Cluster Configuration
  • 2. Data Parallelism
  • 3. Embedding-Sharded Data Parallelism
  • 4. Sync training with unbalanced data across workers.