Spaces:
Sleeping
Sleeping
File size: 1,312 Bytes
23a5053 09f0cd3 d23a6f2 2f2edb0 fdef447 d23a6f2 54bb1a5 0eff587 d23a6f2 54bb1a5 f5a2b6c fdef447 cad50da 27c948e 0eff587 f5a2b6c cad50da fdef447 23a5053 |
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 |
import torch
import torch.nn as nn
import torch.nn.functional as F
class EvoTransformerV22(nn.Module):
def __init__(self, d_model=384, num_heads=6, ffn_dim=1024, num_layers=6, memory_enabled=False):
super().__init__()
self.embedding = nn.Embedding(30522, d_model)
encoder_layer = nn.TransformerEncoderLayer(
d_model=d_model,
nhead=num_heads,
dim_feedforward=ffn_dim,
batch_first=True
)
self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)
self.memory_enabled = memory_enabled
if memory_enabled:
self.memory_token = nn.Parameter(torch.zeros(1, 1, d_model))
self.memory_proj = nn.Linear(d_model, d_model)
self.pooling = nn.AdaptiveAvgPool1d(1)
self.classifier = nn.Sequential(
nn.Linear(d_model, 128),
nn.ReLU(),
nn.Linear(128, 2)
)
def forward(self, input_ids):
x = self.embedding(input_ids)
if self.memory_enabled:
mem = self.memory_token.expand(x.size(0), -1, -1)
x = torch.cat([mem, x], dim=1)
x = self.transformer(x)
x = x.permute(0, 2, 1) # for pooling
x = self.pooling(x).squeeze(-1)
return self.classifier(x)
|