Help: can't run roulette_wheel with DeepSeek model

[ME-Msc]

I am trying to run roulette_wheel with DeepSeek model.

I followed the doc.

The only code I add is

import os
import logfire
from dotenv import load_dotenv

load_dotenv()
logfire.configure()
deepseek_model = OpenAIModel(
    model_name="deepseek-chat",
    base_url="https://api.deepseek.com",
    api_key=os.getenv("OPENAI_API_KEY"),
)

And I changed the model of roulette_agent

roulette_agent = Agent(
    model=deepseek_model,
    deps_type=Deps,
    retries=3,
    result_type=bool,
    system_prompt=(
        "Use the `roulette_wheel` function to determine if the "
        "customer has won based on the number they bet on."
    ),
)

I met an error at async with roulette_agent.run_stream in main().

Expected delta with content, invalid chunk: ChatCompletionChunk(id='f97676fd-771c-44f7-9b2d-e528689fdbc9', choices=[Choice(delta=ChoiceDelta(content=None, function_call=None, refusal=None, role=None, tool_calls=[ChoiceDeltaToolCall(index=0, id='call_0_a7ed393b-31cf-4857-bf10-2199d63240ff', function=ChoiceDeltaToolCallFunction(arguments='', name='roulette_wheel'), type='function')]), finish_reason=None, index=0, logprobs=None)], created=1736505898, model='deepseek-chat', object='chat.completion.chunk', service_tier=None, system_fingerprint='fp_3a5770e1b4', usage=None)

Could anyone tell me why it happened? Did I call the model in the wrong way?
I think @imfing may know, could you help me with a complete example of roulette_wheel?

[ME-Msc]

The logfire shows like below.

[samihamine]

Hi @ME-Msc ,

I believe the error you encountered is related to the issue described in #149. It seems to arise in scenarios where the model is tasked with generating a chunk that includes both a function call and standard content. Specifically, in your case, the ChatCompletionChunk contains a tool_call (indicating a function call to roulette_wheel) but lacks accompanying content or a well-structured response.

At this stage, it appears that OpenAIStreamTextResponse class do not fully handle this particular case—when a chunk contains both a function_call and standard content. This behavior has been identified as a limitation in the current implementation.

From what I understand, there is an ongoing refactor of the class behavior (in this PR #468) that could address this issue and should potentially resolve similar problems where mixed outputs are expected in a single chunk.

[ME-Msc]

@samihamine Thanks for your help! I believe many people have encountered this error, and I will continue to monitor this issue.

Additionally, I would like to suggest that the error message be raised more explicitly and clearly so that newcomers like me can quickly identify the problem.