File size: 1,718 Bytes
73dbe16 8cbf94e 73dbe16 8cbf94e 73dbe16 8cbf94e 73dbe16 8cbf94e 73dbe16 |
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 |
import torch
from torch import nn
class MLP(nn.Module):
def __init__(self,dim):
super().__init__()
self.proj_1 = nn.Linear(dim,dim,bias=False)
self.proj_2 = nn.Linear(dim,dim,bias=False)
self.gelu = nn.GELU()
def forward(self, x):
x = self.proj_1(x)
x = self.gelu(x)
x = self.proj_2(x)
return x
class Context_PReLUBlock(nn.Module):
def __init__(self, d_model, num_tokens):
super().__init__()
self.context_prelu = nn.PReLU(d_model * num_tokens)
self.token_norm = nn.LayerNorm(d_model)
self.local_mapping = MLP(d_model)
def forward(self, x):
residual = x
x = self.token_norm(x)
dim0 = x.shape[0]
dim1 = x.shape[1]
dim2 = x.shape[2]
context = x.reshape([dim0,dim1*dim2])
readout = self.context_prelu(context)
x = readout.reshape([dim0,dim1,dim2])
x = x + residual
residual = x
x = self.local_mapping(x)
out = x + residual
return out
class Context_PReLU(nn.Module):
def __init__(self, d_model,num_tokens, num_layers):
super().__init__()
self.model = nn.Sequential(
*[Context_PReLUBlock(d_model,num_tokens) for _ in range(num_layers)]
)
def forward(self, x):
return self.model(x)
|