[Bug]: Function call exception in some llms

[wangligen1995]

Bug Description

llama_index.core.workflow.errors.WorkflowRuntimeError: Error in step 'run_agent_step': the JSON object must be str, bytes or bytearray, not NoneType

the function not need arguments，gpt-4o return is ''(empty string),some llms return None,When this happens, the workflow ends abnormally.
Although exception is used here, but the err is json.JSONDecodeError not ValueError

filepath:https://github.com/run-llama/llama_index/blob/main/llama-index-integrations/llms/llama-index-llms-openai/llama_index/llms/openai/base.py

Version

0.12.23

Steps to Reproduce

this demo
https://github.com/run-llama/python-agents-tutorial/blob/main/6_multi_agent.py
with other llms

Relevant Logs/Tracbacks

[dosubot]

@wangligen1995 Greetings! I'm here to help you with any bugs, questions, or contributions.

I found a similar issue that was previously reported and resolved. The issue was titled "[Bug]: SlackToolSpec.get_fn_schema_from_fn_name needs to return a schema" and was closed by a pull request that fixed the "fn_schema is None" issue ^[1].

To address the current issue you're facing with NoneType responses leading to a WorkflowRuntimeError, you can adjust the error handling in the base.py file of the llama_index library. Specifically, you can modify the _parse_tool method to check for None before attempting to parse the JSON. Here's a suggested modification:

def _parse_tool(self, tool_call: ToolCall) -> ToolSelection:
    if not isinstance(tool_call, ToolCall):
        raise ValueError("Invalid tool_call object")

    if tool_call.type != "function":
        raise ValueError(f"Unsupported tool call type: {tool_call.type}")

    # Check if tool_call.function.arguments is None
    if tool_call.function.arguments is None:
        argument_dict = {}
    else:
        try:
            argument_dict = parse_partial_json(tool_call.function.arguments)
        except ValueError:
            argument_dict = {}

    return ToolSelection(
        tool_id=tool_call.id,
        tool_name=tool_call.function.name,
        tool_kwargs=argument_dict,
    )

This change ensures that if tool_call.function.arguments is None, it initializes argument_dict as an empty dictionary, thus preventing any NoneType related errors during JSON parsing ^[2].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}