15 lines
631 B
Python
15 lines
631 B
Python
"""evalharness.agent -- evaluation driver for agent benchmarks.
|
|
|
|
NOT a general agent framework: no thinking policies. A message pump that
|
|
lets the model under test act through Environment plugins (bfcl mock today;
|
|
tau2/swe envs land later) and records trajectories for env_reward scorers.
|
|
|
|
from evalharness.agent import drive, BFCLEnvironment
|
|
traj = await drive(adapter, sample, env=BFCLEnvironment())
|
|
"""
|
|
|
|
from .loop import Environment, Trajectory, drive, trajectory_to_prediction
|
|
from .envs.bfcl_mock import BFCLEnvironment
|
|
|
|
__all__ = ['Environment', 'Trajectory', 'drive', 'trajectory_to_prediction', 'BFCLEnvironment']
|