Open
Description
Describe the bug
Hello, I run a simple test for function tools calling with gpt-4o. However, I met a bug "Invalid parameter: messages with role 'tool' must be a response to a preceeding message with 'tool_calls'. " . It seems like the orders of message is wrong. And I have found last year OpenAI updates have introduced a specific requirement that a tool call response must directly follow the message where it was initiated.
Debug information
- Agents SDK version: 0.0.17
- Python 3.10
Repro steps
external_client = AsyncOpenAI(
api_key="xxxx",
base_url="xxxx",
)
set_default_openai_client(external_client)
model = OpenAIChatCompletionsModel(
model="gpt-4o",
openai_client=external_client
)
@function_tool
def get_weather(city: str) -> str:
return f"The weather in {city} is sunny"
agent = Agent(
name="Haiku agent",
instructions="Always respond in haiku form",
model=model,
tools=[get_weather],
)
async def main():
user_input = "What is the weather in Paris?"
result = await Runner.run(
agent,
input=user_input,
)
print(result.final_output())
if __name__ == "__main__":
asyncio.run(main())Expected behavior
I was wondering whether openai agents framework supported this update or not? And I also tried other LLM, doesn't met this bug.
Thank you!