From 21c2b7629b8092c40cba55db4419a789b3b67434 Mon Sep 17 00:00:00 2001 From: Daphne Barretto Date: Tue, 28 Oct 2025 10:44:48 -0700 Subject: [PATCH] Add consistent scores validation (#368) * Add consistent scores validation * revert osworld_run_maestro.py changes --- run.py | 2 +- run_autoglm.py | 2 +- run_autoglm_v.py | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/run.py b/run.py index cce45bc..75148e3 100644 --- a/run.py +++ b/run.py @@ -218,7 +218,7 @@ def test(args: argparse.Namespace, test_all_meta: dict) -> None: f.write("\n") env.close() - logger.info(f"Average score: {sum(scores) / len(scores)}") + logger.info(f"Average score: {sum(scores) / len(scores) if scores else 0}") def get_unfinished( diff --git a/run_autoglm.py b/run_autoglm.py index 8b161c7..5776886 100644 --- a/run_autoglm.py +++ b/run_autoglm.py @@ -457,7 +457,7 @@ def test(args: argparse.Namespace, test_all_meta: dict) -> None: f.write("\n") env.close() - logger.info(f"Average score: {sum(scores) / len(scores)}") + logger.info(f"Average score: {sum(scores) / len(scores) if scores else 0}") def get_unfinished(action_space, use_model, observation_type, result_dir, total_file_json): diff --git a/run_autoglm_v.py b/run_autoglm_v.py index 31794ea..0f95af0 100644 --- a/run_autoglm_v.py +++ b/run_autoglm_v.py @@ -485,7 +485,7 @@ def test(args: argparse.Namespace, test_all_meta: dict) -> None: f.write("\n") env.close() - logger.info(f"Average score: {sum(scores) / len(scores)}") + logger.info(f"Average score: {sum(scores) / len(scores) if scores else 0}") def get_unfinished(action_space, use_model, observation_type, result_dir, total_file_json):