improve gguf-function-calling parser

Author: okaris

llama_cpp/llama_chat_format.py

@@ -4363,7 +4363,7 @@ def gguf_function_calling(
     prompt += "<function_calls>\n"
     if stream:
         return _stream_tool_calls(
-            llama, prompt, tools, tool_name, completion_kwargs, follow_up_gbnf_tool_grammar
+            llama, prompt, tools, tool_name, completion_kwargs, follow_up_gbnf_tool_grammar, message_content
         )
     tool = next((tool for tool in tools if tool["function"]["name"] == tool_name), None)
     completions: List[llama_types.CreateCompletionResponse] = []
@@ -4431,7 +4431,7 @@ def gguf_function_calling(
                 ),
                 "message": {
                     "role": "assistant",
-                    "content": None,
+                    "content": message_content,  # Include message content if present
                     "tool_calls": [
                         {
                             "id": "call_" + f"_{i}_" + tool_name + "_" + completion["id"],