Spaces:
Running
Running
File size: 1,566 Bytes
437e42f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 |
import torch
import torch.nn as nn
import torch.nn.functional as F
# A simple MLP layer
class FeedForwardNetwork(nn.Module):
def __init__(self, input_size, hidden_size, output_size, dropout_rate=0):
super(FeedForwardNetwork, self).__init__()
self.dropout_rate = dropout_rate
self.linear1 = nn.Linear(input_size, hidden_size)
self.linear2 = nn.Linear(hidden_size, output_size)
def forward(self, x):
x_proj = F.dropout(F.relu(self.linear1(x)), p=self.dropout_rate, training=self.training)
x_proj = self.linear2(x_proj)
return x_proj
# Span Prediction for Start Position
class PoolerStartLogits(nn.Module):
def __init__(self, hidden_size, num_classes):
super(PoolerStartLogits, self).__init__()
self.dense = nn.Linear(hidden_size, num_classes)
def forward(self, hidden_states, p_mask=None):
x = self.dense(hidden_states)
return x
# Span Prediction for End Position
class PoolerEndLogits(nn.Module):
def __init__(self, hidden_size, num_classes):
super(PoolerEndLogits, self).__init__()
self.dense_0 = nn.Linear(hidden_size, hidden_size)
self.activation = nn.Tanh()
self.LayerNorm = nn.LayerNorm(hidden_size)
self.dense_1 = nn.Linear(hidden_size, num_classes)
def forward(self, hidden_states, start_positions=None, p_mask=None):
x = self.dense_0(torch.cat([hidden_states, start_positions], dim=-1))
x = self.activation(x)
x = self.LayerNorm(x)
x = self.dense_1(x)
return x
|