bellofils
diff --git a/‎llama_cpp/llama_chat_format.py
Copy file name to clipboardExpand all lines: llama_cpp/llama_chat_format.py
+4Lines changed: 4 additions & 0 deletions b/‎llama_cpp/llama_chat_format.py
Copy file name to clipboardExpand all lines: llama_cpp/llama_chat_format.py
+4Lines changed: 4 additions & 0 deletions
@@ -2088,6 +2088,7 @@ def chatml_function_calling(
         "{% endif %}"
         "{% endif %}"
         "{% endfor %}"
+        "{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
     )
     template_renderer = jinja2.Environment(
         loader=jinja2.BaseLoader(),
@@ -2130,6 +2131,7 @@ def chatml_function_calling(
             messages=messages,
             tools=[],
             tool_calls=None,
+            add_generation_prompt=True,
         )
         if response_format is not None and response_format["type"] == "json_object":
             try:
@@ -2363,6 +2365,7 @@ def _stream_response_to_function_stream(
             messages=messages,
             tools=tools,
             tool_calls=True,
+            add_generation_prompt=True,
         )
         prompt += f"functions.{tool_name}:\n"
         try:
@@ -2420,6 +2423,7 @@ def _stream_response_to_function_stream(
         messages=messages,
         tools=tools,
         tool_calls=True,
+        add_generation_prompt=True,
     )
     completion_or_chunks = llama.create_completion(
         prompt=prompt,