oneflow.one_embedding.MultiTableEmbedding.forward

MultiTableEmbedding.forward(ids, table_ids=None)

Embedding lookup operation

Parameters
  • ids (flow.tensor) – the feature ids

  • table_ids (flow.tensor, optional) – the table_id of each id, must be same shape as ids. There is no need to pass table_ids, if has config only one table or the ids has shape (batch_size, num_tables), and each column’s id belongs to the column_id th table, otherwise, you should pass the tensor_ids.

Returns

the result of embedding lookup

Return type

flow.tensor