openvino.runtime.opset15.embedding_bag_offsets#

openvino.runtime.opset15.embedding_bag_offsets(emb_table: Node | int | float | ndarray, indices: Node | int | float | ndarray, offsets: Node | int | float | ndarray, default_index: Node | int | float | ndarray | None = None, per_sample_weights: Node | int | float | ndarray | None = None, reduction: Literal['sum', 'mean'] = 'sum', name: str | None = None) → Node#

Return a node which performs sums or means of bags of embeddings without the intermediate embeddings.

Parameters:

  • emb_table – Tensor containing the embedding lookup table.

  • indices – 1D Tensor with indices.

  • offsets – 1D Tensor containing the starting index positions of each bag in indices.

  • per_sample_weights – Tensor with weights for each sample.

  • default_index – Scalar containing default index in embedding table to fill empty bags. If unset or set to -1, empty bags will be filled with 0. Reverse indexing using negative indices is not supported.

  • reduction – String to select algorithm used to perform reduction of elements in bag.

  • name – Optional name for output node.

Returns:

The new node performing EmbeddingBagOffsets operation.