forked from hnu202409060624/python
parent
06db13138c
commit
d32ec7f19f
@ -0,0 +1,16 @@
|
|||||||
|
from torch import nn
|
||||||
|
import config
|
||||||
|
class PoswiseFeedForwardNet(nn.Module):
|
||||||
|
def __init__(self):
|
||||||
|
super(PoswiseFeedForwardNet, self).__init__()
|
||||||
|
self.fc = nn.Sequential(
|
||||||
|
nn.Linear(config.input_dim, config.d_ff1, bias=config.bias),
|
||||||
|
nn.ReLU(),
|
||||||
|
nn.Linear(config.d_ff1, config.input_dim, bias=config.bias))
|
||||||
|
|
||||||
|
def forward(self, inputs): # inputs: [batch_size, seq_len, d_model]
|
||||||
|
residual = inputs
|
||||||
|
|
||||||
|
output = self.fc(inputs)
|
||||||
|
|
||||||
|
return nn.LayerNorm(config.input_dim).to(config.device)(output + residual) # [batch_size, seq_len, d_model]
|
Loading…
Reference in new issue