From 772504bda12b161906ea6699494ec062c1a7e41f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?oliver=20k=C3=B6nig?= Date: Wed, 27 May 2026 19:00:51 +0000 Subject: [PATCH] fix(test): widen iter-time steady-state window for short tests MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit For tests that run for ~25 steps (e.g. gpt3_7b_tp1_pp4_memory_speed) the hard-coded steps 30-45 window left both lists empty and produced a NaN median. Select range(5, 21) when the golden run has <=25 steps and keep range(30, 46) otherwise. Signed-off-by: oliver könig --- tests/functional_tests/python_test_utils/common.py | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/tests/functional_tests/python_test_utils/common.py b/tests/functional_tests/python_test_utils/common.py index 57f181f495d..6f528b9a95c 100644 --- a/tests/functional_tests/python_test_utils/common.py +++ b/tests/functional_tests/python_test_utils/common.py @@ -208,9 +208,8 @@ def pipeline( ] if metric_name == "iteration-time": - # Restrict iter-time aggregation to the steady-state window - # (steps 30-45) so the warmup step does not dominate the median. - steady_window = range(30, 46) + max_golden_step = max(golden_value.values.keys()) if golden_value.values else 0 + steady_window = range(5, 21) if max_golden_step <= 25 else range(30, 46) actual_value_list = [ value for value_step, value in actual_values[metric_name].values.items()