Task¶

`gepa.optimize_anything.Task(name: str, seed_candidate: str | dict[str, str] | None = None, objective: str = '', background: str = '', train_set: list[Any] | None = None, val_set: list[Any] | None = None, test_set: list[Any] | None = None)` `dataclass` ¶

A task definition for the optimize_anything API.

Attributes:

Name	Type	Description
`name`	`str`	Unique identifier (e.g. `"circle_packing"`).
`seed_candidate`	`str \| dict[str, str] \| None`	Seed to evolve from — a single text string, or a `{component: text}` dict for multi-component optimization (the `gepa` engine co-optimizes the components; other engines treat the seed as a single text). `None` for seedless mode, where the engine bootstraps the first candidate from `objective` / `background`.
`objective`	`str`	Short goal statement (e.g. "Maximize sum of circle radii"). Surfaced verbatim by every engine as the optimization goal.
`background`	`str`	Long-form context — problem statement, evaluation rules, domain notes. Surfaced verbatim by every engine.
`train_set`	`list[Any] \| None`	Optional training examples (used for optimization). Items are opaque — any object the evaluator understands.
`val_set`	`list[Any] \| None`	Optional validation examples (used for candidate selection).
`test_set`	`list[Any] \| None`	Optional held-out test examples. Held outside the eval server and scored (seed + optimized) outside the budget only when present; skipped entirely when omitted.