Update app.py

app.py

@@ -141,9 +141,6 @@ def build_messages(user_text, profile, blocks):
         {"role": "user", "content": user_text}
     ]
 
-# ----------------------------
-# Generate
-# ----------------------------
 def chat(user_text, model_label, blocks_json):
     data = load_blocks()
     blocks = parse_blocks_editor(blocks_json, data.get("language_blocks", []))
@@ -153,14 +150,14 @@ def chat(user_text, model_label, blocks_json):
 
     messages = build_messages(user_text, data["user_profile"], blocks)
 
-   inputs = tokenizer.apply_chat_template(
-    messages,
-    add_generation_prompt=True,
-    tokenize=True,
-    return_tensors="pt",
-    return_dict=True   # <-- this makes inputs a dict
+    inputs = tokenizer.apply_chat_template(
+        messages,
+        add_generation_prompt=True,
+        tokenize=True,
+        return_tensors="pt",
+        return_dict=True   # ensures inputs is a dict, not just a tensor
     ).to("cpu")
-    
+
     start = time.time()
     with torch.no_grad():
         outputs = model.generate(
@@ -182,6 +179,7 @@ def chat(user_text, model_label, blocks_json):
 
     return gen_text, metrics
 
+
 def parse_blocks_editor(text, fallback):
     if not text or not text.strip():
         return fallback