Tool docstring cannot be longer than 1024 characters in v1 but can be longer in v0

[off6atomic]

Description

This is the code for Mirascope v0:

from dotenv import load_dotenv
from mirascope.openai import OpenAICall, OpenAICallParams

load_dotenv()


def list_design_option_values_sorted_by_sales_metric(
    design_option: str,
    time_range_start: str,
    time_range_end: str,
    sales_metric: str = "TotalUnitsSold",
    asc: bool = False,
):
    """
    Lists pattern design option values sorted by their sales metric in ascending or descending order.
    Top 5 design option values fetched from the Postgres database will be returned along with
    their sales metric.

    Less than 5 design option values will be returned if there are less than 5 values.
    Do not call this function again with the same arguments if you receive less than 5 values.

    Args:
        design_option: The design option to query. MUST be one of the following values:
            - "shoulderSeam"
            - "necklineStyle"
            - "necklineStyle.crewNeck.crewNeckWidth"
            - "necklineStyle.crewNeck.neckFinish"
            - "productLength"
            - "productLength.hip.level"
            - "frontHem"
            - "frontHem.straight.hemFoldBack"
            - "backHem"
            - "backHem.straight.hemFoldBack"
            - "slitDepth"
            - "pocket"
            - "sleeveLength"
            - "sleeveFabric"
            - "sleeveHem"
            - "sleeveHem.regular.sleeveHemFoldBack"
            - "bodyFit"
            - "shoulderFit"
            - "silhouette"
            - "sleeveFit"
        time_range_start: The start of the time range to filter sales by e.g., "2023-01-01 00:00:00+00".
        time_range_end: The end of the time range to filter sales by e.g., "2024-01-01 00:00:00+00".
        sales_metric: Sales metric to sort by. Can be "UnitsSoldFullPrice", "RevenueFullPrice", "UnitsSoldDiscounted", "RevenueDiscounted", "TotalUnitsSold", or "TotalRevenue". Default is "TotalUnitsSold".
        asc: Whether to sort the results in ascending order. Default is False (descending order).

    Raises:
        ValueError: If invalid design_option or sales_metric is provided.
    """
    print(
        "The values are:",
        design_option,
        time_range_start,
        time_range_end,
        sales_metric,
        asc,
    )


class LLM(OpenAICall):
    prompt: str
    prompt_template = "{prompt}"
    call_params = OpenAICallParams(
        model="gpt-4o",
        tools=[list_design_option_values_sorted_by_sales_metric],
        temperature=0.0,
    )


llm = LLM(prompt="call the list tool with random parameter values. don't ask me anything.")
response = llm.call()
if tool := response.tool:
    print(tool.args)
    tool.call()
else:
    print(response.content)

You will notice that the code runs successfully. Here's the output:

{'design_option': 'necklineStyle', 'time_range_start': '2023-01-01 00:00:00+00', 'time_range_end': '2023-12-31 23:59:59+00', 'sales_metric': 'TotalRevenue', 'asc': True}
The values are: necklineStyle 2023-01-01 00:00:00+00 2023-12-31 23:59:59+00 TotalRevenue True

This is the code for Mirascope v1:

from dotenv import load_dotenv
from mirascope.core import openai

load_dotenv()


def list_design_option_values_sorted_by_sales_metric(
    design_option: str,
    time_range_start: str,
    time_range_end: str,
    sales_metric: str = "TotalUnitsSold",
    asc: bool = False,
):
    """
    Lists pattern design option values sorted by their sales metric in ascending or descending order.
    Top 5 design option values fetched from the Postgres database will be returned along with
    their sales metric.

    Less than 5 design option values will be returned if there are less than 5 values.
    Do not call this function again with the same arguments if you receive less than 5 values.

    Args:
        design_option: The design option to query. MUST be one of the following values:
            - "shoulderSeam"
            - "necklineStyle"
            - "necklineStyle.crewNeck.crewNeckWidth"
            - "necklineStyle.crewNeck.neckFinish"
            - "productLength"
            - "productLength.hip.level"
            - "frontHem"
            - "frontHem.straight.hemFoldBack"
            - "backHem"
            - "backHem.straight.hemFoldBack"
            - "slitDepth"
            - "pocket"
            - "sleeveLength"
            - "sleeveFabric"
            - "sleeveHem"
            - "sleeveHem.regular.sleeveHemFoldBack"
            - "bodyFit"
            - "shoulderFit"
            - "silhouette"
            - "sleeveFit"
        time_range_start: The start of the time range to filter sales by e.g., "2023-01-01 00:00:00+00".
        time_range_end: The end of the time range to filter sales by e.g., "2024-01-01 00:00:00+00".
        sales_metric: Sales metric to sort by. Can be "UnitsSoldFullPrice", "RevenueFullPrice", "UnitsSoldDiscounted", "RevenueDiscounted", "TotalUnitsSold", or "TotalRevenue". Default is "TotalUnitsSold".
        asc: Whether to sort the results in ascending order. Default is False (descending order).

    Raises:
        ValueError: If invalid design_option or sales_metric is provided.
    """
    print(
        "The values are:",
        design_option,
        time_range_start,
        time_range_end,
        sales_metric,
        asc,
    )


@openai.call(
    "gpt-4o",
    tools=[list_design_option_values_sorted_by_sales_metric],
    call_params={"temperature": 0},
)
def llm(prompt: str):
    return prompt


response = llm("call the tool with random parameter values. don't ask me anything.")
if tool := response.tool:
    print(tool.args)
    tool.call()
else:
    print(response.content)

Running the code will result in this error:

openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid 'tools[0].function.description': string too long. Expected a string with maximum length 1024, but got a string with length 1628 instead.", 'type': 'invalid_request_error', 'param': 'tools[0].function.description', 'code': 'string_above_max_length'}}

Python, Mirascope & OS Versions, related packages (not required)

Requirements for v0:

openai==1.34.0
pydantic==2.7.1
pydantic-core==2.18.2
mirascope==0.18.2

Requirements for v1:

openai==1.59.7
pydantic==2.10.5
pydantic-core==2.27.2
mirascope==1.15.0

[willbakst]

The 1024 limit is imposed by OpenAI. The reason this worked in v0 and doesn't work in v1 is that in v0 we only used the short/long description and dropped everything else (but in v1 we just use the original docstring).

I've implemented a fix for this that drops args/examples from the tool description so that they don't count against that limit.

The fix is released in v1.16.1!