Update app.py

app.py

@@ -35,13 +35,21 @@ def generate_text(prompt, max_new_tokens=80, temperature=0.7):
     
     with torch.no_grad():
         for _ in range(max_new_tokens):
-            # Pass the signal through the Moiré field
-            logits, _ = model(input_ids)
+            # CROP THE INPUT: Only look at the most recent max_seq_len tokens
+            # so the positional embeddings never go out of bounds (257)
+            cond_input = input_ids[:, -config.max_seq_len:]
+            
+            # Pass the cropped signal through the Moiré field
+            logits, _ = model(cond_input)
+            
+            # Grab the prediction for the last token
             next_token_logits = logits[:, -1, :] / temperature
             
             # Sample the next token
             probs = torch.nn.functional.softmax(next_token_logits, dim=-1)
             next_token = torch.multinomial(probs, num_samples=1)
+            
+            # Append it to the running sequence
             input_ids = torch.cat((input_ids, next_token), dim=1)
             
             # Stop if the field decides the thought is complete