fix: handle non-string function arguments in tool calls and add corresponding warnings

[ekzhu]

Resolves #5257

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 70.42%. Comparing base (fff201f) to head (deaa3f1).
Report is 2 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #5260       +/-   ##
===========================================
- Coverage   83.70%   70.42%   -13.29%     
===========================================
  Files          60      179      +119     
  Lines        2817    11625     +8808     
===========================================
+ Hits         2358     8187     +5829     
- Misses        459     3438     +2979     

Flag	Coverage Δ
unittests	70.42% <100.00%> (-13.29%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lspinheiro]

I found it quite weird how the openai SDK was validating it and converting to dict when Function is a pydantic model but then I realised they have a custom implementation of pydantic BaseModel at openai._models.py.

It could be a bug in their SDK that maybe we should report. Based on the definition and data validation the function should never return a dict.

class Function(BaseModel):
    arguments: str
    """
    The arguments to call the function with, as generated by the model in JSON
    format. Note that the model does not always generate valid JSON, and may
    hallucinate parameters not defined by your function schema. Validate the
    arguments in your code before calling your function.
    """

    name: str
    """The name of the function to call."""

[ekzhu]

Yes. Would you like to report it? Though fixing that will immediately break hugging face users.