Update evo_model.py

evo_model.py

@@ -4,24 +4,27 @@ import torch.nn as nn
 import math
 
 class PositionalEncoding(nn.Module):
-    def __init__(self, d_model, max_len=128):  # ❗ Reverted to 128 to match saved model
+    def __init__(self, d_model, max_len=128):  # Match saved model: [1, 128, 512]
         super().__init__()
         pe = torch.zeros(max_len, d_model)
-        position = torch.arange(0, max_len).unsqueeze(1)
-        div_term = torch.exp(torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))
+        position = torch.arange(0, max_len, dtype=torch.float).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)  # (1, max_len, d_model)
         self.register_buffer('pe', pe)
 
     def forward(self, x):
-        return x + self.pe[:, :x.size(1)]
+        seq_len = x.size(1)
+        if seq_len > self.pe.size(1):
+            raise ValueError(f"Input length {seq_len} exceeds max_len {self.pe.size(1)}")
+        return x + self.pe[:, :seq_len]
 
 class EvoDecoderModel(nn.Module):
     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=128)  # ❗ Match checkpoint shape
+        self.token_embed = nn.Embedding(vocab_size, d_model)       # ✅ matches saved key: token_embed.weight
+        self.pos_encoder = PositionalEncoding(d_model)             # ✅ fixed dimension and safe slicing
         decoder_layer = nn.TransformerDecoderLayer(
             d_model=d_model,
             nhead=nhead,
@@ -30,12 +33,12 @@ 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.lm_head = nn.Linear(d_model, vocab_size)              # ✅ matches saved key: lm_head.weight
 
     def forward(self, input_ids):
-        x = self.token_embed(input_ids)
-        x = self.pos_encoder(x)
+        x = self.token_embed(input_ids)         # (B, T, D)
+        x = self.pos_encoder(x)                 # (B, T, D)
         seq_len = x.size(1)
         mask = torch.triu(torch.ones(seq_len, seq_len, device=x.device), diagonal=1).bool()
         x = self.decoder(x, x, tgt_mask=mask)
-        return self.lm_head(x)
+        return self.lm_head(x)                  # (B, T, V)