Spaces:
Sleeping
Sleeping
File size: 1,680 Bytes
a03a6ca 09f0cd3 cad50da 2f2edb0 84c0256 27c948e 38cb891 27c948e 38cb891 5ed25f6 cad50da 27c948e 38cb891 cad50da 27c948e cad50da 84c0256 38cb891 27c948e 38cb891 84c0256 27c948e |
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 |
import torch
import torch.nn as nn
from torch.nn import TransformerEncoder, TransformerEncoderLayer
class EvoEncoder(nn.Module):
def __init__(self, vocab_size=30522, d_model=512, nhead=8, num_layers=6, dim_feedforward=1024, dropout=0.1):
super(EvoEncoder, self).__init__()
self.embedding = nn.Embedding(vocab_size, d_model)
self.memory_token = nn.Parameter(torch.zeros(1, 1, d_model))
self.positional_encoding = nn.Parameter(torch.zeros(1, 512, d_model))
encoder_layer = TransformerEncoderLayer(d_model=d_model, nhead=nhead,
dim_feedforward=dim_feedforward, dropout=dropout)
self.transformer = TransformerEncoder(encoder_layer, num_layers=num_layers)
self.norm = nn.LayerNorm(d_model)
def forward(self, input_ids):
x = self.embedding(input_ids)
bsz = x.size(0)
# Add memory token
mem_token = self.memory_token.expand(bsz, -1, -1) # [B, 1, D]
x = torch.cat([mem_token, x], dim=1)
x = x + self.positional_encoding[:, :x.size(1), :]
x = self.transformer(x)
x = self.norm(x)
return x[:, 0] # return memory token output
class EvoTransformer(nn.Module):
def __init__(self, vocab_size=30522, d_model=512, nhead=8, num_layers=6,
dim_feedforward=1024, dropout=0.1):
super(EvoTransformer, self).__init__()
self.encoder = EvoEncoder(vocab_size, d_model, nhead, num_layers, dim_feedforward, dropout)
self.classifier = nn.Linear(d_model, 1)
def forward(self, input_ids):
x = self.encoder(input_ids)
return x # shape: [batch, d_model]
|