Class: Secryst::TransformerDecoderLayer

Inherits:
Torch::NN::Module
  • Object
show all
Defined in:
lib/secryst/transformer.rb

Instance Method Summary collapse

Constructor Details

#initialize(d_model, nhead, dim_feedforward: 2048, dropout: 0.1, activation: "relu") ⇒ TransformerDecoderLayer

TransformerDecoderLayer is made up of self-attn, multi-head-attn and feedforward network. This standard decoder layer is based on the paper “Attention Is All You Need”. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems, pages 6000-6010. Users may modify or implement in a different way during application. Args:

d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
activation: the activation function of intermediate layer, relu or gelu (default=relu).
Examples

>>> decoder_layer = TransformerDecoderLayer(512, 8) >>> memory = Torch.rand(10, 32, 512) >>> tgt = Torch.rand(20, 32, 512) >>> out = decoder_layer.call(tgt, memory)



298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
 
# File 'lib/secryst/transformer.rb', line 298

def initialize(d_model, nhead, dim_feedforward: 2048, dropout: 0.1, activation: "relu")
  super()
  @self_attn = MultiheadAttention.new(d_model, nhead, dropout: dropout)
  @multihead_attn = MultiheadAttention.new(d_model, nhead, dropout: dropout)
  # Implementation of Feedforward model
  @linear1 = Torch::NN::Linear.new(d_model, dim_feedforward)
  @dropout = Torch::NN::Dropout.new(p: dropout)
  @linear2 = Torch::NN::Linear.new(dim_feedforward, d_model)

  @norm1 = Torch::NN::LayerNorm.new(d_model)
  @norm2 = Torch::NN::LayerNorm.new(d_model)
  @norm3 = Torch::NN::LayerNorm.new(d_model)
  @dropout1 = Torch::NN::Dropout.new(p: dropout)
  @dropout2 = Torch::NN::Dropout.new(p: dropout)
  @dropout3 = Torch::NN::Dropout.new(p: dropout)

  @activation = _get_activation_fn(activation)
end

Instance Method Details

#forward(tgt, memory, tgt_mask: nil, memory_mask: nil, tgt_key_padding_mask: nil, memory_key_padding_mask: nil) ⇒ Object

Pass the inputs (and mask) through the decoder layer. Args:

tgt: the sequence to the decoder layer (required).
memory: the sequence from the last layer of the encoder (required).
tgt_mask: the mask for the tgt sequence (optional).
memory_mask: the mask for the memory sequence (optional).
tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
memory_key_padding_mask: the mask for the memory keys per batch (optional).

Shape:

see the docs in Transformer class.



327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
 
# File 'lib/secryst/transformer.rb', line 327

def forward(tgt, memory, tgt_mask: nil, memory_mask: nil,
          tgt_key_padding_mask: nil, memory_key_padding_mask: nil)

  tgt2 = @self_attn.call(tgt, tgt, tgt, attn_mask: tgt_mask,
                        key_padding_mask: tgt_key_padding_mask)[0]
  tgt = tgt + @dropout1.call(tgt2)
  tgt = @norm1.call(tgt)
  tgt2 = @multihead_attn.call(tgt, memory, memory, attn_mask: memory_mask,
                             key_padding_mask: memory_key_padding_mask)[0]
  tgt = tgt + @dropout2.call(tgt2)
  tgt = @norm2.call(tgt)
  tgt2 = @linear2.call(@dropout.call(@activation.call(@linear1.call(tgt))))
  tgt = tgt + @dropout3.call(tgt2)
  tgt = @norm3.call(tgt)
  return tgt
end