File size: 3,016 Bytes
f4e579f |
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 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 |
import torch
from torch import nn
class FeedForward(nn.Module):
def __init__(self, dim, hidden_dim, dropout):
super().__init__()
self.net = nn.Sequential(
nn.Linear(dim, hidden_dim),
nn.GELU(),
nn.Dropout(dropout),
nn.Linear(hidden_dim, dim),
nn.Dropout(dropout)
)
def forward(self, x):
return self.net(x)
class ContextualizerBlock(nn.Module):
def __init__(self, d_model,d_ffn,dropout,num_tokens):
super().__init__()
self.context_proj = nn.Linear(d_model,d_model)
self.mlp = FeedForward(d_model,d_ffn,dropout)
self.norm = nn.LayerNorm(d_model)
self.upsample = nn.Upsample(scale_factor=num_tokens,mode='nearest')
self.downsample = nn.Upsample(scale_factor= 1/num_tokens, mode='nearest')
def forward(self, x):
res = x
x = self.norm(x)
context = x
dim0 = context.shape[0]
dim1 = context.shape[1]
dim2 = context.shape[2]
context = context.reshape([dim0,1,dim1*dim2])
context = self.downsample(context)
context = context.reshape([dim0,dim2])
context = self.context_proj(context)
context = context.reshape([dim0,1,dim2])
context = self.upsample(context)
context = context.reshape([dim0,dim1,dim2])
x = context
x = x + res
res = x
x = self.norm(x)
x = self.mlp(x)
out = x + res
return out
return
class MixerGatingUnit(nn.Module):
def __init__(self,d_model,d_ffn,dropout,num_tokens):
super().__init__()
self.Mixer = ContextualizerBlock(d_model,d_ffn,dropout,num_tokens)
self.proj = nn.Linear(d_model,d_model)
def forward(self, x):
u, v = x, x
u = self.proj(u)
v = self.Mixer(v)
out = u * v
return out
class ContextualizerNiNBlock(nn.Module):
def __init__(self, d_model,d_ffn,dropout,num_tokens):
super().__init__()
self.norm = nn.LayerNorm(d_model)
self.mgu = MixerGatingUnit(d_model,d_ffn,dropout,num_tokens)
self.ffn = FeedForward(d_model,d_ffn,dropout)
def forward(self, x):
residual = x
x = self.norm(x)
x = self.mgu(x)
x = x + residual
residual = x
x = self.norm(x)
x = self.ffn(x)
out = x + residual
return out
class ContextualizerNiN(nn.Module):
def __init__(self, d_model, d_ffn, num_layers,dropout,num_tokens):
super().__init__()
self.model = nn.Sequential(
*[ContextualizerNiNBlock(d_model,d_ffn,dropout,num_tokens) for _ in range(num_layers)],
)
def forward(self, x):
x = self.model(x)
return x
|