Automatically convert types if models send strings to tools expecting integers/floats

llm --functions '
def multiply(x: int, y: int) -> int:
    """Multiply two numbers."""
    return x * y
' 'what is 34234 * 213345' -m qwen3:4b

I've noticed that qwen3 4B sometimes ignores the integer types and sends strings instead. It would be neat if LLM spotted that and converted them before handing them to the function, based on the signature.

Start by reproducing the shown llm --functions command with the Qwen model, then trace the tool-call path that hands model arguments to the Python function. Determine where the function signature is available and add conversion of string values to declared integer or float types before invocation. Done means the example succeeds with string arguments while existing tool calls remain supported.