Fix choosing the right transforms

[ahuang11]

1. Shortens prompts and redundancy since longer context lengths take longer to process and ruins the smaller models
2. Renames provides -> fulfills (not sure if matters; was just desperate to try things to get it work)
3. removes decision and uses llm_validator instead
4. bumps temperature (not sure if matters)

[philippjfr]

Looks good, I think I would stick with provides over fulfill. Other than that and the linting issues I'm happy to merge.

[ahuang11]

Dropped the llm_validator because it was using OpenAI underneath and it didn't work with LlamaCpp right out of the box without manual patching https://github.com/jxnl/instructor/blob/cea534fd2280371d2778e0f043d3fe557cc7bc7e/instructor/dsl/validators.py#L74

Also the typing of it was somewhat complex, so I replaced it something similar that captures the underlying logic and provided field descriptions thru FieldInfo when creating the transform model; still works successfully.

        transform_model = create_model(
            "Transform",
            summary_of_query=(
                str,
                FieldInfo(
                    description="A summary of the query and whether you think a transform is needed."
                ),
            ),
            transform_required=(
                bool,
                FieldInfo(
                    description="Whether a transform is required for the query; sometimes the user may not need a transform.",
                    default=True,
                ),
            ),

[philippjfr]

Looks good, just make pre-commit happy and I'll merge.

[philippjfr]

Okay, let's fix tests on main. I'll merge this PR though.

[github-actions[bot]]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.