[Bug]: Tool description exceeds maximum length of 1024 characters. Please shorten your description or move it to the prompt.

[danerlt]

Bug Description

When I use the workflow and use the tool, it prompts me: Tool description exceeds maximum length of 1024 characters. Please shorten your description or move it to the prompt.

Version

0.12.19

Steps to Reproduce

When I run the workflow and use a tool with a very long description, an error occurs：Tool description exceeds maximum length of 1024 characters. Please shorten your description or move it to the prompt.

I see that to_openai_tool has a skip_length_check parameter that can skip the length check.

    def to_openai_tool(self, skip_length_check: bool = False) -> Dict[str, Any]:
        """To OpenAI tool."""
        if not skip_length_check and len(self.description) > 1024:
            raise ValueError(
                "Tool description exceeds maximum length of 1024 characters. "
                "Please shorten your description or move it to the prompt."
            )
        return {
            "type": "function",
            "function": {
                "name": self.name,
                "description": self.description,
                "parameters": self.get_parameters_dict(),
            },
        }

However, in OpenAI's _prepare_chat_with_tools method, when calling to_openai_tool, this parameter is not passed.

class OpenAI(FunctionCallingLLM):

    def _prepare_chat_with_tools(
        self,
        tools: Sequence["BaseTool"],
        user_msg: Optional[Union[str, ChatMessage]] = None,
        chat_history: Optional[List[ChatMessage]] = None,
        verbose: bool = False,
        allow_parallel_tool_calls: bool = False,
        tool_choice: Union[str, dict] = "auto",
        strict: Optional[bool] = None,
        **kwargs: Any,
    ) -> Dict[str, Any]:
        """Predict and call the tool."""
        tool_specs = [tool.metadata.to_openai_tool() for tool in tools]

        # if strict is passed in, use, else default to the class-level attribute, else default to True`
        if strict is not None:
            strict = strict
        else:
            strict = self.strict

        if self.metadata.is_function_calling_model:
            for tool_spec in tool_specs:
                if tool_spec["type"] == "function":
                    tool_spec["function"]["strict"] = strict
                    # in current openai 1.40.0 it is always false.
                    tool_spec["function"]["parameters"]["additionalProperties"] = False

        if isinstance(user_msg, str):
            user_msg = ChatMessage(role=MessageRole.USER, content=user_msg)

        messages = chat_history or []
        if user_msg:
            messages.append(user_msg)

        return {
            "messages": messages,
            "tools": tool_specs or None,
            "tool_choice": resolve_tool_choice(tool_choice) if tool_specs else None,
            **kwargs,
        }

I hope that when calling achat_with_tools, skip_length_check can be passed in.

Relevant Logs/Tracbacks

Traceback (most recent call last):
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\core\workflow\workflow.py", line 270, in _task
    new_ev = await instrumented_step(**kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\core\instrumentation\dispatcher.py", line 368, in async_wrapper
    result = await func(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\src\job_assistant\task\agent\agent.py", line 209, in handle_llm_input
    response = await self.llm.achat_with_tools(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\core\llms\function_calling.py", line 65, in achat_with_tools
    chat_kwargs = self._prepare_chat_with_tools(
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\core\instrumentation\dispatcher.py", line 322, in wrapper
    result = func(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\llms\openai\base.py", line 906, in _prepare_chat_with_tools
    tool_specs = [tool.metadata.to_openai_tool() for tool in tools]
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\incremental-learning\api\.venv\Lib\site-packages\llama_index\core\tools\types.py", line 77, in to_openai_tool
    raise ValueError(
ValueError: Tool description exceeds maximum length of 1024 characters. Please shorten your description or move it to the prompt.

[dosubot]

The _prepare_chat_with_tools method in the OpenAI class already includes the skip_length_check parameter when calling to_openai_tool. The method should look like this:

tool_specs = [
    tool.metadata.to_openai_tool(skip_length_check=True) for tool in tools
]

This line of code ensures that the skip_length_check=True argument is passed to the to_openai_tool method for each tool in the tools list, allowing you to bypass the length check for tool descriptions ^[1]. If this is not reflected in your current version, you might need to update your code to include this parameter.

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}

[logan-markewich]

Are you using openai? It will actually break their api if that request is made. Openai has this limitation

[danerlt]

@logan-markewich No, the LLM I use is of the OpenAILike class.