Gemini changes added 8

app.py

@@ -52,6 +52,42 @@ def load_model():
 
 tokenizer, model, model_name = load_model()
 
+
+# --- Start: Apply CORRECTED Chat Template from File (if Granite) ---
+if "granite" in model_name.lower():
+    template_filename = "granite3.3_2b_chat_template.jinja" # Use the new filename
+    applied_template = False
+    try:
+        # Assuming the template file is in the same directory as app.py (project root)
+        print(f"Attempting to load corrected chat template from: {template_filename}")
+        with open(template_filename, "r", encoding="utf-8") as f:
+            custom_chat_template_content = f.read()
+
+        # Assign the loaded template content
+        tokenizer.chat_template = custom_chat_template_content
+        applied_template = True
+        print(f"✅ Loaded and applied corrected chat template from: {template_filename}")
+
+    except FileNotFoundError:
+        print(f"⚠️ WARNING: Corrected template file '{template_filename}' not found.")
+    except Exception as e:
+        print(f"❌ ERROR reading corrected template file '{template_filename}': {e}")
+
+    # Fallback / Verification print
+    if not applied_template:
+         print("Falling back to tokenizer's default built-in template (which might be incorrect).")
+    print("--- Final Chat Template Being Used ---")
+    if hasattr(tokenizer, 'chat_template') and tokenizer.chat_template:
+        print(tokenizer.chat_template) # Print the template actually being used
+    else:
+        print("Tokenizer does not have a chat_template attribute or it is empty.")
+    print("------------------------------------")
+
+else:
+    print("Model is not Granite, using default chat template.")
+# --- End: Apply CORRECTED Chat Template from File ---
+
+
 # --- Start: Print Loaded Chat Template ---
 print("--- Tokenizer's Loaded Chat Template ---")
 # Check if the attribute exists before printing
@@ -168,11 +204,15 @@ def chat(message, history, hotel_id):
         for role, content in history_tuples:
             messages.append({"role": role, "content": content})
 
+        # Set meta data (annotations)which influences the bot behaviour in the controls json
+        controls = {"length":"short","originality": "abstractive"}
+
         # Apply the template
         input_text = tokenizer.apply_chat_template(
             messages,
             tokenize=False,
-            add_generation_prompt=True
+            add_generation_prompt=True,
+            controls=controls
         )
 
         print("--- Granite Templated Input ---")

granite3.3_2b_chat_template.jinja

@@ -50,15 +50,17 @@ Finally, after the response is written, include a numbered list of sentences fro
          {{- '<|end_of_text|>
 ' }}
               {%- endfor %}
- {%- endif %}
- {%- for message in loop_messages %}
-     {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>
-' }}
-     {%- if loop.last and add_generation_prompt %}
-         {{- '<|start_of_role|>assistant' }}
-             {%- if controls %}
-                 {{- ' ' + controls | tojson()}}
-             {%- endif %}
-         {{- '<|end_of_role|>' }}
-     {%- endif %}
- {%- endfor %}
+ {%- endif %} 
+ {# Find this loop near the end of the template file #}
+{%- for message in loop_messages %}
+    {%- if message['role'] != 'document' %} {# <<< ADD THIS LINE #}
+        {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>\n' }}
+        {%- if loop.last and add_generation_prompt %}
+            {{- '<|start_of_role|>assistant' }}
+                {%- if controls %}
+                    {{- ' ' + controls | tojson()}}
+                {%- endif %}
+            {{- '&lt;|end_of_role|&gt;' }}
+        {%- endif %}
+    {%- endif %} {# <<< ADD THIS LINE #}
+{%- endfor %}