need enhancement for LLM response of function calling embedded in markdown

[1WorldCapture]

Scenario: function calling Steps:

1. run cookbook under "cookbook\llms\ollama\tools\app.py" using "streamlit run app.py"
2. select llama3 model
3. ask question about stock price of some company, like APPLE or GOOGLE

Problem: Sometimes LLM responses raw JSON text, and sometimes it embeds the reponse in markdown, like

{
....
}

or

{
.....
}

Either case could not be handled correctly in current code.

So is it possible or necessary to enhance the cookcode to handle such case? thanks.

[ashpreetbedi]

yup working on this :)

[ju1987yetchung]

I have same problem, I think it is because one parameter should be chosen: tool_choice <Union[str, Dict[str, Any]]>. On the document it is said that if there is tool defined, it is set to "auto" as default value. this means the AI robot can choose from answering a message or using a tool. So, under this conditon, from the documentation, this parameter seemly should be specified, but the pattern is defined blurry, hard to write right code.

[jacobweiss2305]

@ju1987yetchung @1WorldCapture

Please try our new ollama agent class:

phidata\cookbook\providers\ollama

"""Run `pip install yfinance` to install dependencies."""

from phi.agent import Agent
from phi.model.ollama import Ollama
from phi.tools.yfinance import YFinanceTools

agent = Agent(
    model=Ollama(id="llama3.2"),
    tools=[YFinanceTools(stock_price=True, analyst_recommendations=True, stock_fundamentals=True)],
    show_tool_calls=True,
    description="You are an investment analyst that researches stocks and helps users make informed decisions.",
    instructions=["Use tables to display data where possible."],
    markdown=True,
)

# agent.print_response("Share the NVDA stock price and analyst recommendations", stream=True)
agent.print_response("Summarize fundamentals for TSLA", stream=True)

Let us know how it works!