Update evo_model.py

evo_model.py

@@ -1,29 +1,32 @@
-import torch
 import torch.nn as nn
 from torch.nn import TransformerEncoder, TransformerEncoderLayer
 
-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.embedding = nn.Embedding(vocab_size, d_model)
-        self.memory_token = nn.Parameter(torch.zeros(1, 1, 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.memory_proj = nn.Linear(d_model, d_model)
+class EvoEncoder(nn.Module):
+    def __init__(self, d_model=512, nhead=8, dim_feedforward=1024, num_layers=6, dropout=0.1):
+        super().__init__()
+        self.embedding = nn.Embedding(30522, d_model)
+        self.positional_encoding = nn.Parameter(torch.zeros(1, 512, d_model))
+        encoder_layers = TransformerEncoderLayer(d_model, nhead, dim_feedforward, dropout)
+        self.transformer = TransformerEncoder(encoder_layers, num_layers)
         self.norm = nn.LayerNorm(d_model)
-        
-        self.classifier = nn.Linear(d_model, 1)  # Matches saved model: output is a single logit
+        self.memory_proj = nn.Linear(d_model, d_model)
+        self.memory_token = nn.Parameter(torch.zeros(1, 1, d_model))
 
     def forward(self, input_ids):
         x = self.embedding(input_ids)
-
+        x += self.positional_encoding[:, :x.size(1)]
         memory_token = self.memory_token.expand(x.size(0), -1, -1)
         x = torch.cat([memory_token, x], dim=1)
-
         x = self.transformer(x)
         x = self.norm(x)
+        return self.memory_proj(x[:, 0, :])
 
-        memory_output = self.memory_proj(x[:, 0])
-        return memory_output
+class EvoTransformer(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.encoder = EvoEncoder()
+        self.classifier = nn.Linear(512, 1)
+
+    def forward(self, input_ids):
+        x = self.encoder(input_ids)
+        return x