make_litellm_lm

gepa.optimize_anything.make_litellm_lm(model_name: str, **kwargs: Any) -> LanguageModel

Convert a LiteLLM model name string to a :class:LanguageModel callable.

The returned callable conforms to the LanguageModel protocol and accepts a plain str prompt, a list[dict] chat-messages list, or a multimodal messages list (with content arrays containing images).

Uses :class:gepa.lm.LM which handles reasoning model detection (o1/o3/o4/gpt-5), retries with exponential backoff, truncation warnings, and drop_params=True for cross-model compatibility.

Parameters:

Name	Type	Description	Default
model_name	str	LiteLLM model identifier (e.g. "openai/gpt-5").	required
**kwargs	Any	Extra keyword arguments forwarded to litellm.completion (e.g. reasoning_effort="high", temperature=0.7).	{}

Source code in gepa/optimize_anything.py

def make_litellm_lm(model_name: str, **kwargs: Any) -> LanguageModel:
    """Convert a LiteLLM model name string to a :class:`LanguageModel` callable.

    The returned callable conforms to the ``LanguageModel`` protocol and
    accepts a plain ``str`` prompt, a ``list[dict]`` chat-messages list, or
    a multimodal messages list (with content arrays containing images).

    Uses :class:`gepa.lm.LM` which handles reasoning model detection
    (o1/o3/o4/gpt-5), retries with exponential backoff, truncation
    warnings, and ``drop_params=True`` for cross-model compatibility.

    Args:
        model_name: LiteLLM model identifier (e.g. ``"openai/gpt-5"``).
        **kwargs: Extra keyword arguments forwarded to ``litellm.completion``
            (e.g. ``reasoning_effort="high"``, ``temperature=0.7``).
    """
    from gepa.lm import LM

    return LM(model_name, **kwargs)