OCI llama model does not return tool response

[jmalbarranORA]

I'm running the simple LangGraph test in Getting Started

This simple test create a simple (hard-coded) tool to answer about the weather. When I call the sample with Cohere model cohere.command-latestit runs find.

But when I try the Llama model meta.llama-3.3-70b-instruct, it calls the tool, but does not integrate the response in the answer.

This is my code

chat = ChatOCIGenAI(
    auth_type="API_KEY",
    auth_profile=os.getenv("OCI_PROFILE"),
    model_id=os.getenv("OCI_GENAI_MODEL_ID"),
    service_endpoint=os.getenv("OCI_GENAI_SERVICE_ENDPOINT"),
    compartment_id=os.getenv("OCI_COMPARTMENT_OCID"),
    model_kwargs={"temperature": 1, "max_tokens": 500},
)

messages = [  
    SystemMessage(content="You are an AI assistant."),  
    # AIMessage(content="Hi there human!"),  
    HumanMessage(content="What is the weather in SF?"),  
    ]  


agent = create_react_agent(
    model=chat,  
    tools=[get_weather],  
    prompt="You are a helpful assistant"  
)

# Run the agent
response = agent.invoke({"messages": messages})

log_response(response, logging.DEBUG)

# Print last message as response
print(response["messages"][-1].content)

This is the answer using Cohere:

2025-08-31 18:49:48,415 - DEBUG - tool: It's always sunny in SF!
2025-08-31 18:49:48,415 - DEBUG - ai: The weather in SF is always sunny!
The weather in SF is always sunny!

And this the answer using Llama

2025-08-31 18:43:44,568 - DEBUG - tool: It's always sunny in SF!
2025-08-31 18:43:44,568 - DEBUG - ai: I hope this is what you were looking for. let me know if I can help with anything else.
I hope this is what you were looking for. let me know if I can help with anything else.

Notice the debug lines. The tools returns the same (and correct) answer in both case, is the LLM which is not returning the answer

[fede-kamel]

#49