|
from typing import Type |
|
|
|
from torch import nn |
|
|
|
|
|
|
|
|
|
class MLPBlock(nn.Module): |
|
def __init__( |
|
self, |
|
input_dim: int, |
|
hidden_dim: int, |
|
output_dim: int, |
|
num_layers: int, |
|
act: Type[nn.Module], |
|
) -> None: |
|
super().__init__() |
|
self.num_layers = num_layers |
|
h = [hidden_dim] * (num_layers - 1) |
|
self.layers = nn.ModuleList( |
|
nn.Sequential(nn.Linear(n, k), act()) |
|
for n, k in zip([input_dim] + h, [hidden_dim] * num_layers) |
|
) |
|
self.fc = nn.Linear(hidden_dim, output_dim) |
|
|
|
def forward(self, x): |
|
for layer in self.layers: |
|
x = layer(x) |
|
return self.fc(x) |
|
|