messages with role 'tool' must be a response to a preceeding message with 'tool_calls

[brianpham93]

I'm encountering this error 'message': \"Invalid parameter: messages with role 'tool' must be a response to a preceeding message with 'tool_calls from both Azure OpenAI and OpenAI API chat completion endpoints.
It looks like this is due to executor executing multiple functions at once rather than doing it one by one, after getting the function's result. Is there anyway to force that only 1 function could be executed at a time to ensure the order of messages ?

Engineer (to chat_manager):

***** Suggested tool call (call_lSD9ogZbpcKvUUa09Yp9takm): scrape_example *****
Arguments: 
{"url": "https://www.example.com/comments/9hmflv/shabu_sai_wagyu_worth_it/.json"}
******************************************************************************
***** Suggested tool call (call_SZdGCSSiAU5FH50C8MNzTaof): scrape_example *****
Arguments: 
{"url": "https://www.example.com/comments/wndmlz/sukiya_vs_shabu_sai/.json"}
******************************************************************************
***** Suggested tool call (call_QSVSlsZxYqGGDXfJWNGJTgeI): scrape_example *****
Arguments: 
{"url": "https://www.example.com/many_calories_in_hotpotshabu_shabu/.json"}
******************************************************************************

--------------------------------------------------------------------------------

Next speaker: Executor


>>>>>>>> EXECUTING FUNCTION scrape_example...
Scraping website: https://www.example.com/comments/9hmflv/shabu_sai_wagyu_worth_it/.json

>>>>>>>> EXECUTING FUNCTION scrape_example...
Scraping website: https://www.example.com/comments/wndmlz/sukiya_vs_shabu_sai/.json

>>>>>>>> EXECUTING FUNCTION scrape_example...
Scraping website: https://www.example.com/many_calories_in_hotpotshabu_shabu/.json
Executor (to chat_manager):

Executor (to chat_manager):

***** Response from calling tool (call_lSD9ogZbpcKvUUa09Yp9takm) *****
["same thoughts, definitely worth trying out at least once", "hahahha I like their tonkotsu broth if I'm spelling it right, the spicy one legit taste like soy sauce", "Can you share with us which are the nice soups?", "Tonkotsu and sukiya for me! ", "Tonkotsu and sukiya for me! "]
**********************************************************************

--------------------------------------------------------------------------------
Executor (to chat_manager):

***** Response from calling tool (call_SZdGCSSiAU5FH50C8MNzTaof) *****
[]
**********************************************************************

--------------------------------------------------------------------------------
Executor (to chat_manager):

***** Response from calling tool (call_QSVSlsZxYqGGDXfJWNGJTgeI) *****
[]
**********************************************************************

[rshoemake-launch]

I'm having the same exact problem though I wasn't just a week ago or so.

[rshoemake-launch]

I'm having the same exact problem though I wasn't just a week ago or so.

This was with pyautogen 0.2.32 and 0.2.31 thought the later one worked about 1.5 weeks ago.

[jordanbean-msft]

I'm seeing this too. In my case, I have defined 1 set of agents, but am trying to call them over multiple different chats in parallel using asyncio.gather()

[rshoemake-launch]

I just figured out my issue. I had dialed down the following numbers too low, and apparently it was cutting off what was needed.
This was the previous value which cause me the issue: transforms.MessageHistoryLimiter(max_messages=25).
This is what I ended up with:
transforms=[
transforms.MessageHistoryLimiter(max_messages=50),
transforms.MessageTokenLimiter(max_tokens_per_message=2000),
]

[its-arsalann]

@rshoemake-launch, yes, the quick fix is to increase the max_messages but there is no known minimum limit. So, we can't set it for general cases.