Refactor torch.nn.Linear or pointwise_conv

Refactor torch.nn.LayerNorm to meet 4-D dataflow.

The ContextGraph is modified from Aho-Corasick which is mainly a Trie with a fail arc for each node. See https://en.wikipedia.org/wiki/Aho%E2%

Construct an Swish object.

Positionwise feed forward layer. FeedForward are appied on each position of the sequence. The output dim is same with the input dim. Arg

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

Refactor wenet/transformer/positionwise_feed_forward.py::PositionwiseFeedForward

Refactor torch.nn.Identity(). For inserting BPU node whose input == output.

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

The state in ContextGraph

Conv2d operator for VALID mode padding.

Label-smoothing loss. In a standard CE loss, the label's data distribution is: [0,1,2] -> [ [1.0, 0.0, 0.0], [0.0, 1.0, 0

A LayerDrop implementation based on :class:`torch.nn.ModuleList`. We refresh the choice of which layers to drop every time we iterate ov

Positionwise feed forward layer. Args: idim (int): Input dimenstion. hidden_units (int): The number of hidden units. drop

For CTC prefix beam search

Utility class for measuring steps/second.

Streamming Text line

Base class of Transfomer decoder module. Args: vocab_size: output dim encoder_output_size: dimension of attention attentio

Refactor wenet/transformer/ctc.py::CTC

Refactor wenet/transformer/encoder.py::ConformerEncoder

Refactor wenet/transformer/encoder_layer.py::ConformerEncoderLayer

Refactor wenet/transformer/subsampling.py::Conv2dSubsampling8 NOTE(xcsong): Only support pos_enc_class == NoPositionalEncoding

Refactor wenet/transformer/convolution.py::ConvolutionModule NOTE(xcsong): Only suport use_layer_norm == False

Refactor wenet/transformer/cmvn.py::GlobalCMVN

Refactor wenet/transformer/attention.py::MultiHeadedAttention NOTE(xcsong): Only support attention_class == MultiHeadedAttention, we do n

Branchformer encoder layer module. Args: size (int): model dimension attn: standard self-attention or efficient attention, option

Collate function for AudioDataset

Collate function for AudioDataset

ConvolutionModule in Conformer model.

Convolutional Spatial Gating Unit (CSGU).

Single decoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

Depthwise Convolutional 2D subsampling (to 1/4 length). Args: idim (int): Input dimension. odim (int): Output dimensi

A dummy multihead attention for Paraformer befroe cross attention

E-Branchformer encoder layer module. Args: size (int): model dimension attn: standard self-attention or efficient attention

Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer and "Generating Long Sequ

Paraformer subsampling

Sinusoids position encoding used in paraformer.encoder

Relative positional encoding module. See : Appendix B in https://arxiv.org/abs/1901.02860 Args: d_model (int): Embedding dimension.

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

Transformer encoder module.

The WarmupLR scheduler This scheduler is almost same as NoamLR Scheduler except for following difference: NoamLR: lr = optimizer

Paraformer has a decoder3

Conv1d subsampling module. Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rate (float): Dropout

Conv1d subsampling module. Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rate (float): Dropout

Conv1d subsampling module. Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rate (float): Dropout

Conv1d subsampling module. Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rate (float): Dropout

CTC-attention hybrid Encoder-Decoder model

Base class of Transfomer decoder module. Args: vocab_size: output dim encoder_output_size: dimension of attention attentio