Tool calling doesn't return tool_calls results in streaming mode

[a1exwang]

Repro steps:

1. Install AI Toolkit, download qwen2.5 7b cpu model
2. Run the testing scripts
3. response.choices[0].delta.tool_calls is always empty. But content contains xml tool call info.

Expected:
By OpenAI documentation, the chat completion API tool call response should be in response.choices[0].delta.tool_calls. Documentation.

Testing script

import os
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:5272/v1",
    api_key="unused",
)

# Define the function schema
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather in a given location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g., San Francisco, CA"
                    },
                    "unit": {
                        "type": "string",
                        "enum": ["celsius", "fahrenheit"],
                        "description": "The temperature unit to use"
                    }
                },
                "required": ["location"]
            }
        }
    }
]

# Make the initial streaming request
stream = client.chat.completions.create(
    model='qwen2.5-7b-instruct-generic-cpu:4',
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What's the weather like in San Francisco?"}
    ],
    tools=tools,
    tool_choice="auto",
    stream=True
)

print("Streaming Response:")
for chunk in stream:
    print(chunk.choices[0].model_dump_json())

Version:

Foundry Local in AI Toolkit: 0.26.4
0.7.122
1.23.0-dev-20250918-0932-9c85c395b0
0.9.1-rc1
0.10.0
0.8.4

[natke]

You’ll need to set tool_choice="required" rather than auto

[natke]

We have reproduced this issue for streaming, even with tool_choice="required"