Tools

At a glance

The post indicates that the get_parameters_dict function is problematic. The comments suggest that the tool call arguments should be more descriptive, with a structured schema for parameters. Community members discuss issues with the current implementation, such as the limited character count in the main tool description and the lack of proper fine-tuning. They also explore alternative approaches like using fn_schema, but note that it doesn't accept a dictionary or JSON format. The discussion involves various related concepts and functions, but there is no explicitly marked answer.

zzensin

get_parameters_dict is problemmatic

10 comments

LLogan M

What's wrong with it? Do open a PR if you spot an issue 🙏

zzensin

meh. maybe

zzensin

it's that tool call arguments are supposed to be pretty descriptive with like:

Plain Text

"parameters": {
                    "type": "object",
                    "properties": {
                        "name": {
                            "type": "string",
                            "description": "The name of the thing.",
                        },
"required": ["name"]

zzensin

docstrings can provide it but the current llamaindex implementation just dumps it all in the main tool description

zzensin

which is limited in how many characters it can hold, and isn't how gpt is fine-tuned

zzensin

maybe I'll feel better about a PR after sleep

zzensin

wait maybe I was just relying on that when I could set fn_schema

zzensin

fn_schema doesn't take a dict or json. It wants, something...

zzensin

yeah it's not doing the fun thing

zzensin

so let's dump some tags here...
.schema()
deprecated
model_json_schema
to_tool_list()
parameters.items()
get_parameters_dict()
to_openai_tool()
ToolOutput

DefaultToolFnSchema
input

Add a reply

Find answers from the community

Tools