Update evo_model.py

evo_model.py

@@ -1,27 +1,27 @@
-# evo_model.py — FIXED for checkpoint with pos_encoder.pe
+# evo_model.py — EvoDecoder model with extended positional encoding
 import torch
 import torch.nn as nn
 import math
 
 class PositionalEncoding(nn.Module):
-    def __init__(self, d_model, max_len=128):
+    def __init__(self, d_model, max_len=512):  # Increased from 128 to 512
         super().__init__()
         pe = torch.zeros(max_len, d_model)
         position = torch.arange(0, max_len, dtype=torch.float32).unsqueeze(1)
         div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
         pe[:, 0::2] = torch.sin(position * div_term)
         pe[:, 1::2] = torch.cos(position * div_term)
-        pe = pe.unsqueeze(0)  # shape [1, max_len, d_model]
-        self.pe = nn.Parameter(pe, requires_grad=False)
+        pe = pe.unsqueeze(0)  # shape: [1, max_len, d_model]
+        self.register_buffer('pe', pe)
 
     def forward(self, x):
         return x + self.pe[:, :x.size(1)]
 
 class EvoDecoderModel(nn.Module):
-    def __init__(self, vocab_size, d_model=512, nhead=8, num_layers=6, dim_feedforward=2048, dropout=0.1, max_len=128):
+    def __init__(self, vocab_size, d_model=512, nhead=8, num_layers=6, dim_feedforward=2048, dropout=0.1):
         super().__init__()
-        self.token_embed = nn.Embedding(vocab_size, d_model)
-        self.pos_encoder = PositionalEncoding(d_model, max_len)
+        self.embedding = nn.Embedding(vocab_size, d_model)
+        self.pos_encoder = PositionalEncoding(d_model)
         decoder_layer = nn.TransformerDecoderLayer(
             d_model=d_model,
             nhead=nhead,
@@ -30,12 +30,13 @@ class EvoDecoderModel(nn.Module):
             batch_first=True
         )
         self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=num_layers)
-        self.lm_head = nn.Linear(d_model, vocab_size)
+        self.linear = nn.Linear(d_model, vocab_size)
 
     def forward(self, input_ids):
-        x = self.token_embed(input_ids)
-        x = self.pos_encoder(x)
-        seq_len = x.size(1)
-        mask = torch.triu(torch.ones(seq_len, seq_len, device=x.device), diagonal=1).bool()
-        output = self.decoder(x, x, tgt_mask=mask)
-        return self.lm_head(output)
+        embedded = self.embedding(input_ids)
+        embedded = self.pos_encoder(embedded)
+        seq_len = embedded.size(1)
+        mask = torch.triu(torch.ones(seq_len, seq_len, device=embedded.device), diagonal=1).bool()
+        output = self.decoder(embedded, embedded, tgt_mask=mask)
+        logits = self.linear(output)
+        return logits