How initialize weights of a `torch.nn.Transformer` module? - neural-network

Question

I am using a vanilla transformer architecture from the "Attention Is All You Need" paper for a sequence-to-sequence task. As shown in the following code.
Assuming that I would like to use the torch.nn.init.kaiming_uniform_ initialization method, how would one go about initializing the weights of the nn.Transformer ?
Is it necessary to use a custom encoder and decoder class in order for that to happen?
import torch
import torch.nn as nn
import math
# helper Module that adds positional encoding to the token embedding to introduce a notion of word order.
import torch.nn as nn
class PositionalEncoding(nn.Module):
def __init__(self,
emb_size: int,
dropout: float,
maxlen: int = 20):
super(PositionalEncoding, self).__init__()
den = torch.exp(- torch.arange(0, emb_size, 2)* math.log(10000) / emb_size)
pos = torch.arange(0, maxlen).reshape(maxlen, 1)
pos_embedding = torch.zeros((maxlen, emb_size))
pos_embedding[:, 0::2] = torch.sin(pos * den)
pos_embedding[:, 1::2] = torch.cos(pos * den)
pos_embedding = pos_embedding.unsqueeze(-2)
self.dropout = nn.Dropout(dropout)
self.register_buffer('pos_embedding', pos_embedding)
def forward(self, token_embedding: Tensor):
return self.dropout(token_embedding + self.pos_embedding[:token_embedding.size(0), :])
# helper Module to convert tensor of input indices into corresponding tensor of token embeddings
class TokenEmbedding(nn.Module):
def __init__(self, vocab_size: int, emb_size):
super(TokenEmbedding, self).__init__()
self.embedding = nn.Embedding(vocab_size, emb_size)
# Initialize weights with He initialization
self.embedding.weight = nn.init.kaiming_uniform_(self.embedding.weight)
self.emb_size = emb_size
def forward(self, tokens: Tensor):
return self.embedding(tokens.long()) * math.sqrt(self.emb_size)
# Seq2Seq Network
class Transformer(nn.Module):
def __init__(self,
src_vocab_size: int,
tgt_vocab_size: int,
num_encoder_layers: int = 1,
num_decoder_layers: int = 1,
emb_size: int = 300,
nhead: int = 3,
dim_feedforward: int = 512,
dropout: float = 0.1,
activation_function='relu'):
super().__init__()
self.src_tok_emb = TokenEmbedding(src_vocab_size, emb_size)
self.tgt_tok_emb = TokenEmbedding(tgt_vocab_size, emb_size)
self.positional_encoding = PositionalEncoding(
emb_size, dropout=dropout)
self.transformer = torch.nn.Transformer(d_model=emb_size,
nhead=nhead,
num_encoder_layers=num_encoder_layers,
num_decoder_layers=num_decoder_layers,
dim_feedforward=dim_feedforward,
dropout=dropout,
batch_first=True,
activation=activation_function)
self.generator = nn.Linear(emb_size, tgt_vocab_size)
def init_weights(self):
self.generator.weight = nn.init.kaiming_uniform_(self.generator.weight)
def forward(self,
src: Tensor,
trg: Tensor,
src_mask: Tensor,
tgt_mask: Tensor,
src_padding_mask: Tensor,
tgt_padding_mask: Tensor,
memory_key_padding_mask: Tensor):
# the .permute() is necessary since the positional-encoder expects tensors to be of shape
# (seq_len, batch_size, emb_length)
src_emb = self.positional_encoding(self.src_tok_emb(src).permute(1,0,2)).permute(1,0,2)
tgt_emb = self.positional_encoding(self.tgt_tok_emb(trg).permute(1,0,2)).permute(1,0,2)
outs = self.transformer(src_emb, tgt_emb, src_mask, tgt_mask, None,
src_padding_mask, tgt_padding_mask, memory_key_padding_mask)
return self.generator(outs)
def encode(self, src: Tensor, src_mask: Tensor):
return self.transformer.encoder(self.positional_encoding(
self.src_tok_emb(src)), src_mask)
def decode(self, tgt: Tensor, memory: Tensor, tgt_mask: Tensor):
return self.transformer.decoder(self.positional_encoding(
self.tgt_tok_emb(tgt)), memory,
tgt_mask)