Update evo_model.py

evo_model.py

@@ -2,12 +2,16 @@ 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):
+class EvoEncoder(nn.Module):
+    def __init__(self, d_model=384, num_heads=6, ffn_dim=1024, num_layers=6, memory_enabled=True):
         super().__init__()
         self.embedding = nn.Embedding(30522, d_model)
         self.memory_enabled = memory_enabled
-        self.memory_token = nn.Parameter(torch.zeros(1, 1, d_model)) if memory_enabled else None
+        if memory_enabled:
+            self.memory_proj = nn.Linear(d_model, d_model)
+            self.memory_token = nn.Parameter(torch.zeros(1, 1, d_model))
+        else:
+            self.memory_token = None
 
         encoder_layer = nn.TransformerEncoderLayer(
             d_model=d_model,
@@ -16,13 +20,6 @@ class EvoTransformerV22(nn.Module):
             batch_first=True
         )
         self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)
-        self.pool = nn.AdaptiveAvgPool1d(1)
-
-        self.classifier = nn.Sequential(
-            nn.Linear(d_model, 128),
-            nn.ReLU(),
-            nn.Linear(128, 2)  # Binary classification
-        )
 
     def forward(self, input_ids):
         x = self.embedding(input_ids)
@@ -32,5 +29,21 @@ class EvoTransformerV22(nn.Module):
             x = torch.cat([mem, x], dim=1)
 
         x = self.transformer(x)
+        return x
+
+
+class EvoTransformerV22(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.encoder = EvoEncoder(memory_enabled=True)
+        self.pool = nn.AdaptiveAvgPool1d(1)
+        self.classifier = nn.Sequential(
+            nn.Linear(384, 128),
+            nn.ReLU(),
+            nn.Linear(128, 2)
+        )
+
+    def forward(self, input_ids):
+        x = self.encoder(input_ids)
         x = self.pool(x.transpose(1, 2)).squeeze(-1)
         return self.classifier(x)