More Advanced Handling of When Tools are available

[njbrake]

Hi! Maybe I am missing something but is there more of a description available about why this logic exists? This does not work with AWS Sagemaker either, since tools are not supported by Sagemaker so the get_litellm_params liteLLM function throws an Exception about unsupported parameters.

docetl/docetl/operations/utils.py

Line 762 in a58d7a7

and ("ollama" in model or "azure/gpt-4o-mini" in model)

Maybe a short term hack is to add "sagemaker" to this if statement, but I'm not sure if other things would need to be adjusted as well?

[shreyashankar]

Thanks for making the issue! We've just been adding models ad-hoc, when users complain that tool use significantly degrades the output quality (or is entirely unsupported).

A hacky fix is to put sagemaker in that if statement (but you can only have single-key output schemas for operations unfortunately). Feel free to make a PR if adding a "sagemaker" check works :-)

Separately, we are exploring using outlines; as in #127