Update app.py

app.py

@@ -12,6 +12,8 @@ from llama_cpp_agent import MessagesFormatterType
 from llama_cpp_agent.providers import LlamaCppPythonProvider
 from llama_cpp_agent.chat_history import BasicChatHistory
 from llama_cpp_agent.chat_history.messages import Roles
+from llama_cpp_agent.chat_history.messages import Roles
+from llama_cpp_agent.messages_formatter import MessagesFormatter, PromptMarkers
 import gradio as gr
 from huggingface_hub import hf_hub_download
 from typing import List, Tuple
@@ -33,6 +35,27 @@ hf_hub_download(
     local_dir="./models",
 )
 
+
+# Define the prompt markers for Gemma 3
+gemma_3_prompt_markers = {
+    Roles.system: PromptMarkers("", "\n"),  # System prompt should be included within user message
+    Roles.user: PromptMarkers("<start_of_turn>user\n", "<end_of_turn>\n"),
+    Roles.assistant: PromptMarkers("<start_of_turn>model\n", "<end_of_turn>\n"),
+    Roles.tool: PromptMarkers("", ""),  # If you need tool support
+}
+
+# Create the formatter
+gemma_3_formatter = MessagesFormatter(
+    pre_prompt="",  # No pre-prompt
+    prompt_markers=gemma_3_prompt_markers,
+    include_sys_prompt_in_first_user_message=True,  # Include system prompt in first user message
+    default_stop_sequences=["<end_of_turn>", "<start_of_turn>"],
+    strip_prompt=False,  # Don't strip whitespace from the prompt
+    bos_token="<bos>",  # Beginning of sequence token for Gemma 3
+    eos_token="<eos>",  # End of sequence token for Gemma 3
+)
+
+
 # Set the title and description
 title = "Gemma Llama.cpp"
 description = """Gemma 3 is a family of lightweight, multimodal open models that offers advanced capabilities like large context windows and multilingual support, enabling diverse applications on various devices."""
@@ -92,7 +115,7 @@ def respond(
         agent = LlamaCppAgent(
             provider,
             system_prompt=f"{system_message}",
-            predefined_messages_formatter_type=MessagesFormatterType.GEMMA_2,
+            predefined_messages_formatter_type=gemma_3_formatter,
             debug_output=True,
         )
 
@@ -141,7 +164,11 @@ def respond(
 # Create a chat interface
 demo = gr.ChatInterface(
     respond,
-    examples=[["What is the capital of France?"], ["Tell me something about artificial intelligence."], ["What is gravity?"]],
+    examples=[
+        ["What is the capital of France?"],
+        ["Tell me something about artificial intelligence."],
+        ["What is gravity?"],
+    ],
     additional_inputs_accordion=gr.Accordion(
         label="⚙️ Parameters", open=False, render=False
     ),