From febed9be8ddf72f95ed8661b0bf7e877caeedd19 Mon Sep 17 00:00:00 2001 From: Ryan McCormick Date: Fri, 25 Oct 2024 12:48:27 -0700 Subject: [PATCH] Disable triton profile tests due to genai-perf bug in 24.09 - should be restored in 24.10 --- tests/utils.py | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/tests/utils.py b/tests/utils.py index 6f42251..88bd5ee 100644 --- a/tests/utils.py +++ b/tests/utils.py @@ -73,6 +73,15 @@ def _infer(model, prompt=None, protocol=None): run(args) def _profile(model, backend): + # FIXME: WAR for genai-perf bug in 24.09, remove in 24.10 + import genai_perf + + if genai_perf.__version__ == "0.0.6dev": + print( + "[WARNING] Skipping call to 'triton profile' due to known issue in genai-perf" + ) + return + args = ["profile", "-m", model, "--backend", backend] # NOTE: With default parameters, genai-perf may take upwards of 1m30s or 2m to run, # so limit the genai-perf run with --request-count to reduce time for testing purposes.