mindspore.nn.TransformerEncoder

class mindspore.nn.TransformerEncoder(encoder_layer, num_layers, norm=None)[source]

Transformer Encoder module with multi-layer stacked of TransformerEncoderLayer, including multihead attention and feedforward layer. Users can build the BERT(https://arxiv.org/abs/1810.04805) model with corresponding parameters.

Parameters

encoder_layer (Cell) – An instance of the mindspore.nn.TransformerEncoderLayer class.
num_layers (int) – The number of encoder-layers in the encoder.
norm (Cell, optional) – The layer normalization module. Default: None.

Inputs:

src (Tensor): The sequence to the encoder. For unbatched input, the shape is \((S, E)\) ; otherwise if batch_first=False in TransformerEncoderLayer, the shape is \((S, N, E)\) and if batch_first=True , the shape is \((S, N, E)\), where \((S)\) is the source sequence length, \((N)\) is the batch number and \((E)\) is the feature number. Supported types: float16, float32, float64.
src_mask (Tensor, optional): The mask of the src sequence. The shape is \((S, S)\) or \((N*nhead, S, S)\) , where nhead is the arguent in TransformerDecoderLayer. Supported types: float16, float32, float64, bool. Default: None.
src_key_padding_mask (Tensor, optional): the mask of the src keys per batch. The shape is \((S)\) for unbatched input, otherwise \((N, S)\) . Supported types: float16, float32, float64, bool. Default: None.

Outputs:

Tensor. The shape and dtype of Tensor is the same with src .

Raises: AssertionError – If the input argument src_key_padding_mask is not bool or floating types.

Supported Platforms:: Ascend GPU CPU

Examples

>>> import mindspore as ms
>>> import numpy as np
>>> encoder_layer = ms.nn.TransformerEncoderLayer(d_model=512, nhead=8)
>>> transformer_encoder = ms.nn.TransformerEncoder(encoder_layer, num_layers=6)
>>> src = ms.Tensor(np.random.rand(10, 32, 512), ms.float32)
>>> out = transformer_encoder(src)
>>> print(out.shape)
(10, 32, 512)