[Feature]: litellm.enable_json_schema_validation be converted to a per-call parameter instead of module level setting.

[achpalaman]

The Feature

Move enable_json_schema_validation from litellm.enable_json_schema_validation to a parameter provided to the completion( ) api.

Motivation, pitch

Currently, we must set litellm.enable_json_schema_validation = True to enable json schema validation.
This is very useful, because if schema validation fails, then LiteLLM will retry or fallback to other models.

We exploit this property by first trying with Claude, and falling back to GPT Structured Outputs if Claude fails to produce a schema matching the spec.

However, there are times when we want to disable this validation (when we have our own error correcting mechanism, or custom parser.) -- Unfortunately, since this isn't a parameter to the completion( ) API, it means we either have it for all requests, or None.

Are you a ML Ops Team?

Yes

Twitter / LinkedIn details

No response