Update evo_model.py

evo_model.py

@@ -1,9 +1,8 @@
-import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
 class EvoEncoder(nn.Module):
-    def __init__(self, d_model=384, num_heads=6, ffn_dim=1024, num_layers=6, memory_enabled=False):
+    def __init__(self, d_model=512, num_heads=8, ffn_dim=1024, num_layers=6, memory_enabled=True):
         super().__init__()
         self.embedding = nn.Embedding(30522, d_model)
         encoder_layer = nn.TransformerEncoderLayer(
@@ -14,18 +13,24 @@ class EvoEncoder(nn.Module):
         )
         self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)
         self.memory_enabled = memory_enabled
+        if memory_enabled:
+            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)
+        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)
-        return x[:, 0]  # first token
+        return x[:, 0]  # Return memory token or first token
 
 class EvoTransformer(nn.Module):
-    def __init__(self, d_model=384, num_heads=6, ffn_dim=1024, num_layers=6, num_classes=2, memory_enabled=False):
+    def __init__(self, d_model=512, num_heads=8, ffn_dim=1024, num_layers=6, num_classes=1, memory_enabled=True):
         super().__init__()
         self.encoder = EvoEncoder(d_model, num_heads, ffn_dim, num_layers, memory_enabled)
         self.classifier = nn.Linear(d_model, num_classes)
 
     def forward(self, input_ids):
         x = self.encoder(input_ids)
-        return self.classifier(x)
+        return self.classifier(x)