| """EJBench evaluation harness: load tasks, prompt a model, score answers. |
| |
| Scorers return a value in [0, 1] per task; the headline metric is the macro mean over task |
| families. Multi-step tasks run through ``StoredDataEnv`` (tool calls under a budget; the |
| hidden state is never shown to the model). |
| """ |
|
|
| from .data import FAMILIES, load_tasks |
| from .env import StoredDataEnv |
| from .parsing import extract_json |
| from .scoring import score_task |
|
|
| __all__ = ["FAMILIES", "StoredDataEnv", "extract_json", "load_tasks", "score_task"] |
|
|
