Update app_chat.py

app_chat.py

@@ -86,7 +86,7 @@ def generate(
         use_cache = True,
         repetition_penalty=repetition_penalty,
         stopping_criteria = stopping_criteria,
-        attention_mask = torch.ones_like(tokenized_chat),  # Add this
+        attention_mask = torch.ones_like(input_ids),  # Add this
         position_ids = None,
         kv_last_layer = None,
     )