HugeCTR API Documentation

• Python Interface
  • About the HugeCTR Python Interface
  • High-level Training API
    • Solver
      • CreateSolver method
    • AsyncParam
      • AsyncParam class
    • HybridEmbeddingParam
      • HybridEmbeddingParam class
    • DataReaderParams
      • DataReaderParams class
    • Dataset formats
      • Data Files
      • File List
      • Raw
      • Parquet
    • OptParamsPy
      • CreateOptimizer method
  • Layers
    • Model
      • Model class
      • add method
      • compile method
      • fit method
      • summary method
      • graph_to_json method
      • construct_from_json method
      • load_dense_weights method
      • load_dense_optimizer_states method
      • load_sparse_weights method
      • load_sparse_optimizer_states method
      • freeze_dense method
      • freeze_embedding method
      • unfreeze_dense method
      • unfreeze_embedding method
      • reset_learning_rate_scheduler method
      • set_source method
  • Low-level Training API
    • LearningRateScheduler
      • get_next method
    • DataReader
      • set_source method
      • is_eof method
    • Model
      • get_learning_rate_scheduler method
      • get_data_reader_train method
      • get_data_reader_eval method
      • start_data_reading method
      • set_learning_rate method
      • train method
      • get_current_loss method
      • eval method
      • get_eval_metrics method
      • save_params_to_files method
      • check_out_tensor method
  • Inference API
    • InferenceParams
      • InferenceParams class
    • InferenceModel
      • InferenceModel class
      • predict method
      • evaluate method
      • check_out_tensor method
  • Data Generator API
    • DataGeneratorParams class
    • DataGenerator
      • DataGenerator class
      • generate method
  • Data Source API
    • DataSourceParams class
• Layer Classes and Methods
  • Input Layer
  • Sparse Embedding
  • Embedding Types Detail
    • DistributedSlotSparseEmbeddingHash Layer
    • LocalizedSlotSparseEmbeddingHash Layer
    • LocalizedSlotSparseEmbeddingOneHot Layer
  • Dense Layers
  • Dense Layers Usage
    • FullyConnected Layer
    • MLP Layer
    • MultiCross Layer
    • FmOrder2 Layer
    • WeightMultiply Layer
    • ElementwiseMultiply Layer
    • BatchNorm Layer
    • LayerNorm Layer
    • Concat Layer
    • Reshape Layer
    • Select Layer
    • Slice Layer
    • Dropout Layer
    • ELU Layer
    • ReLU Layer
    • Sigmoid Layer
    • Interaction Layer
    • Add Layer
    • ReduceSum Layer
      • GRU Layer
      • PReLUDice Layer
      • Scale Layer
      • FusedReshapeConcat Layer
      • FusedReshapeConcatGeneral Layer
      • Softmax Layer
      • Sub Layer
      • ReduceMean Layer
      • MatrixMutiply Layer
      • MultiHeadAttention Layer
      • SequenceMask Layer
      • Gather Layer
    • BinaryCrossEntropyLoss
    • CrossEntropyLoss
    • MultiCrossEntropyLoss
  • Embedding Collection
    • About the HugeCTR embedding collection
    • Overview of using the HugeCTR embedding collection
    • Known Limitations
    • EmbeddingTableConfig
    • EmbeddingCollectionConfig
      • embedding_lookup method
      • shard method