INNER CODE UNIT · Python
grade_workspace
wgpsec/AboutSecurity · scripts/grade_eval.py:171
def grade_workspace(workspace: Path, evals_data: dict = None, model: str = None):
"""Grade all runs in a workspace directory."""
run_dirs = sorted(workspace.rglob("run-*/"))
if not run_dirs:
print(f"No run directories found in {workspace}")
return
print(f"Grading {len(run_dirs)} runs in {workspace}")
for run_dir in run_dirs:
if not (run_dir / "output.txt").exists():
continue
rel = run_dir.relative_to(workspace)
print(f" ▸ {rel}...", end=" ", flush=True)
grading = grade_run_dir(run_dir, evals_data, model=model)