EmbCacheEmbeddingCollection

torch.optim.Adagrad: Adagrad optimizer.
torch.optim.Adam: Adam optimizer.
torch.optim.SGD: SGD optimizer.

Creates a single-server table object with hash mapping and multi-level cache.

class EmbCacheEmbeddingCollection:
     def __init__(**kwargs):

Parameter	Data Type	Mandatory/Optional	Description
tables	List[EmbCacheEmbeddingConfig\|EmbeddingConfig]	Mandatory	List of sparse table configurations. The list length range is [1, 10000].
world_size	int	Mandatory	Size of world_size for distributed training. The value range is [1, 10000].
batch_size	int	Mandatory	Batch size. The value range is [1, 102400].
multi_hot_sizes	List[int]	Mandatory	List of multi-hot encoding sizes of each feature. The length of this parameter list must be the same as that of the tables list. The value range is [1, 10000]. The value range of the multi-hot encoding size in the list is [1, 102400].
need_indices	bool	Optional	Whether an index is required. The default value is False.
need_accumulate_offset	bool	Optional	Whether to accumulate offsets. The default value is True.
device	torch.device	Optional	The default value is CPU. The compute device is the same as that of EmbCacheEmbeddingBagCollection.
embedding_optimizer_cls	Type[torch.optim.Optimizer]	Optional	Embedding optimizer type. The default value is torch.optim.Adagrad. The options are as follows: torch.optim.Adagrad: Adagrad optimizer. torch.optim.Adam: Adam optimizer. torch.optim.SGD: SGD optimizer.

Parent topic: Table Creation APIs