[BUG] OpenAIServerModel doesn't work with vLLM serve properly

[ErokhinVi]

Describe the bug
I have deployed the Qwen 2.5 32B model using vllm serve. I created an OpenAIServerModel and am trying to run an agent on it. However, I get an error on the first and subsequent steps

Code to reproduce the error
`model = OpenAIServerModel(
model_id="large",
api_base=llm_api_url,
api_key=os.getenv("LLM_API_KEY"),
temperature=0.2,
)

agent = CodeAgent(
tools=[],
add_base_tools=False,
model=model,
max_steps=10,
additional_authorized_imports=[
"json",
"pandas",
"sqlite3",
],
)

agent.run("How to make pizza ?")
`

Error logs (if any)
Error in generating model output:
Error code: 500 - {'message': 'Internal server error'}

Expected behavior
I expect that i have no error :))

Packages version:
smolagents==1.9.2

Additional context
I figured out how to fix this. The error occurs due to the flatten_messages_as_text parameter in the models.py script. If it is manually set to True by default, the issue is resolved. However, I haven't found a way to set it via kwargs or other methods :(