API reference

TorchRec provides a comprehensive set of APIs for building and deploying recommendation systems at scale. This section covers the core components including data types for sparse and dense features, modular building blocks for recommendation models, distributed training with model parallelism, and optimized inference capabilities.

• Data Types
  • JaggedTensor
  • KeyedJaggedTensor
  • KeyedTensor
• Modules
  • EmbeddingBagConfig
  • EmbeddingConfig
  • BaseEmbeddingConfig
  • EmbeddingBagCollection
  • EmbeddingCollection
• Planner
  • ShardingPlan
  • EmbeddingShardingPlanner
  • EmbeddingEnumerator
  • GreedyPerfPartitioner
  • HeuristicalStorageReservation
  • GreedyProposer
  • EmbeddingPerfEstimator
  • EmbeddingStorageEstimator
• Model Parallel
  • DistributedModelParallel
• Inference
  • quantize_inference_model()
  • shard_quant_model()