There is something wrong with ollama

[lixy910915]

when I do ollama test in 02_llm_powered_agents file, in Implementing Function Calling Capabilities part, I cannot get the final result like this

"🤖 Function-Enabled Agent wants to call a function:

Function: get_weather
Parameters:
location: Tokyo

⚙️ Function result:

{'temperature': 26, 'conditions': 'Sunny', 'humidity': 70, 'wind_speed': 5, 'unit': 'celsius', 'location': 'Tokyo'}
Parsing function call from text: Based on the weather data provided:

The current weather in Tokyo is:

• Temperature: 26°C
• Conditions: Sunny
• Humidity: 70%
• Wind Speed: 5 km/h

It's a warm and sunny day in Tokyo with moderate humidity. The light wind speed of 5 km/h should make for pleasant conditions.

🤖 Function-Enabled Agent final response:

Based on the weather data provided:

The current weather in Tokyo is:

• Temperature: 26°C
• Conditions: Sunny
• Humidity: 70%
• Wind Speed: 5 km/h

It's a warm and sunny day in Tokyo with moderate humidity. The light wind speed of 5 km/h should make for pleasant conditions."

I got final result like this:
"
Function-enabled OpenAI agent created successfully!
🔍 Testing Function-Enabled Agent function calling...
INFO:httpx:HTTP Request: POST http://127.0.0.1:11434/v1/chat/completions "HTTP/1.1 200 OK"

🤖 Function-Enabled Agent wants to call a function:

Function: get_weather
Parameters:
location: Tokyo

⚙️ Function result:

{'temperature': 26, 'conditions': 'Sunny', 'humidity': 70, 'wind_speed': 5, 'unit': 'celsius', 'location': 'Tokyo'}
INFO:httpx:HTTP Request: POST http://127.0.0.1:11434/v1/chat/completions "HTTP/1.1 200 OK"

🤖 Function-Enabled Agent final response:

......

To provide the weather information, I need to know the location you're interested in. Could you please specify a city or region?
"
I used qwq:latest with ollama

[themanojdesai]

@x42en could you please check this .?

[x42en]

Yes sure, please @lixy910915 could you share me some code example so I can try with local models ?

[lixy910915]

Hi, here is the code I used:

def get_weather(location, unit="celsius"):
    """Simulate getting weather data for a location."""
    # In a real application, this would call a weather API
    # For this example, we'll use mock data
    weather_data = {
        "New York": {"temperature": 22, "conditions": "Partly Cloudy", "humidity": 65, "wind_speed": 8},
        "London": {"temperature": 18, "conditions": "Rainy", "humidity": 80, "wind_speed": 12},
        "Tokyo": {"temperature": 26, "conditions": "Sunny", "humidity": 70, "wind_speed": 5},
        "Sydney": {"temperature": 28, "conditions": "Clear", "humidity": 55, "wind_speed": 10},
        "Paris": {"temperature": 20, "conditions": "Cloudy", "humidity": 60, "wind_speed": 7}
    }
    
    # Get the weather data for the location (default to New York if not found)
    location_data = weather_data.get(location, weather_data["New York"])
    
    # Convert temperature if necessary
    if unit.lower() == "fahrenheit":
        location_data["temperature"] = location_data["temperature"] * 9/5 + 32
        location_data["unit"] = "fahrenheit"
    else:
        location_data["unit"] = "celsius"
    
    # Add the location to the data
    location_data["location"] = location
    
    return location_data

# Define the function schema for OpenAI
weather_function = {
    "name": "get_weather",
    "description": "Get the current weather for a location",
    "parameters": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The city name, e.g. New York"
            },
            "unit": {
                "type": "string",
                "enum": ["celsius", "fahrenheit"],
                "description": "The temperature unit"
            }
        },
        "required": ["location"]
    }
}

# Create a function-enabled OpenAI agent (if API key is available)
function_agent = OllamaA2AServer(
    api_url="http://127.0.0.1:11434",
    model=
    "qwq:latest",
    temperature=0.7,
    system_prompt="You are a helpful assistant that can provide weather information.",
    functions=[weather_function]
)
print("Function-enabled OpenAI agent created successfully!")

# Now let's create a function to test the function calling capability
def test_function_calling(agent, prompt, agent_name="Agent"):
    """Test an agent's function calling capability with a prompt."""
    print(f"🔍 Testing {agent_name} function calling...")
    
    # Create a message with the prompt
    message = Message(
        content=TextContent(text=prompt),
        role=MessageRole.USER
    )
    
    # Get a response from the agent
    response = agent.handle_message(message)
    
    # Check if the response is a function call
    if response.content.type == "function_call":
        print(f"\n🤖 {agent_name} wants to call a function:\n")
        print(f"Function: {response.content.name}")
        print("Parameters:")
        for param in response.content.parameters:
            print(f"  {param["name"]}: {param["value"]}")
        
        # Execute the function
        if response.content.name == "get_weather":
            params = {p["name"]: p["value"] for p in response.content.parameters}
            result = get_weather(**params)
            
            # Create a function response
            function_response = Message(
                content=FunctionResponseContent(
                    name="get_weather",
                    response=result
                ),
                role=MessageRole.AGENT,
                parent_message_id=response.message_id,
                conversation_id=response.conversation_id
            )
            
            print(f"\n⚙️ Function result:\n")
            print(result)

            # print(f"\n⚙️ Function response:\n")
            # print(function_response)
            
            # Get the final response from the agent
            final_response = agent.handle_message(function_response)
            
            print(f"\n🤖 {agent_name} final response:\n")
            print(final_response.content.text)
            # print(final_response)
            
            return final_response
    else:
        print(f"\n🤖 {agent_name} response (no function call):\n")
        print(response.content.text)
        
    print("\n" + "-"*80 + "\n")
    
    return response

# Test the function calling capability (if API key is available)

# Test with a prompt that should trigger function calling
function_response = test_function_calling(
    function_agent,
    "What's the weather like in Tokyo right now?",
    "Function-Enabled Agent"
)

# Test with another location
# function_response2 = test_function_calling(
#     function_agent,
#     "I'm planning a trip to London. What's the weather there?",
#     "Function-Enabled Agent"
# ))

[x42en]

Ok I found the issue. It seems Ollama requires a function_response role of tool rather than function in lines 139 and 196 of OpenAIA2AServer.

I could either create a PR with:

• a conditional role definition
• use tool

Could someone with OpenAI account double-check the impact of using tool role ?

---

Here is the output using qwq on Ollama once role has been changed:

$ poetry run ./test_a2a.py 
Function-enabled OpenAI agent created successfully!
🔍 Testing Function-Enabled Agent function calling...
INFO:httpx:HTTP Request: POST http://ollama.cydream.lab:11434/v1/chat/completions "HTTP/1.1 200 OK"

🤖 Function-Enabled Agent wants to call a function:

Function: get_weather
Parameters:
  location: Tokyo
  unit: celsius

⚙️ Function result:

{'temperature': 26, 'conditions': 'Sunny', 'humidity': 70, 'wind_speed': 5, 'unit': 'celsius', 'location': 'Tokyo'}
INFO:httpx:HTTP Request: POST http://ollama.cydream.lab:11434/v1/chat/completions "HTTP/1.1 200 OK"

🤖 Function-Enabled Agent final response:

<think>
Okay, the user provided a weather response for Tokyo. Let me check what they need. The original query was to get the weather for Tokyo in Celsius. The tool response has all the details: temperature 26°C, sunny, 70% humidity, wind speed 5 km/h. I should present this info clearly. Maybe start with the location and unit, then list the temperature, conditions, humidity, and wind speed. Keep it friendly and straightforward. Make sure to mention the unit again since they specified Celsius. Alright, time to format the answer.
</think>

Here's the current weather in Tokyo (Celsius):

- **Temperature:** 26°C  
- **Conditions:** Sunny  
- **Humidity:** 70%  
- **Wind Speed:** 5 km/h  

Let me know if you need further details! ☀️

[lixy910915]

OK, Thanks very much! Will you update OllamaA2AServer file separately? I see you used OpenAIA2AServer functions in OllamaA2AServer, I'm not sure is there any other issues if we use OllamaA2AServer in other situations

[x42en]

OK, Thanks very much! Will you update OllamaA2AServer file separately? I see you used OpenAIA2AServer functions in OllamaA2AServer, I'm not sure is there any other issues if we use OllamaA2AServer in other situations

As Ollama claim to have an OpenAI compatible API using /v1/ subpath (ref) I take advantage of the existing openai dependency. If any other error is identified I'll consider re-implement the OllamaA2AServer using ollama python sdk.

But everything is working by my side for what I need to do so it seems good enough to me 😄 If you identify any other problem please open a new issue. Until then I need an OpenAI API key to double check the impact of using tool rather than function so we can fix accordingly.

If I can't find any solution until wednesday I'll push a fix using conditional role definition so it won't break anything.