From f4b28f239848db9f12121e1d760ef204b128e0be Mon Sep 17 00:00:00 2001 From: rein yang <73573651+R2-Y@users.noreply.github.com> Date: Tue, 12 May 2026 22:10:10 +0800 Subject: [PATCH] [CI] update daily omni min accuracy (#3536) Signed-off-by: rein yang --- .../e2e/accuracy/qwen3_omni/run_qwen_omni_acc_benchmark.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/tests/e2e/accuracy/qwen3_omni/run_qwen_omni_acc_benchmark.py b/tests/e2e/accuracy/qwen3_omni/run_qwen_omni_acc_benchmark.py index 7fb71b28d..b0dc2af2d 100644 --- a/tests/e2e/accuracy/qwen3_omni/run_qwen_omni_acc_benchmark.py +++ b/tests/e2e/accuracy/qwen3_omni/run_qwen_omni_acc_benchmark.py @@ -8,7 +8,7 @@ L4-style smoke runs against an already-running server. The pytest wrappers in H100 / MI325) because they launch the live Omni server inside the test. 1. **Daily-Omni** — MCQ accuracy fields in the saved JSON (``daily_omni_accuracy``, …); by default the - run **fails** if accuracy is strictly below **0.69** (``--min-daily-omni-accuracy`` / ``ACC_BENCH_MIN_DAILY_OMNI_ACCURACY``). + run **fails** if accuracy is strictly below **0.67** (``--min-daily-omni-accuracy`` / ``ACC_BENCH_MIN_DAILY_OMNI_ACCURACY``). 2. **Seed-TTS** — ``seed-tts-eval``-style metrics when ``--seed-tts-wer-eval`` is used (WER / SIM / UTMOS keys from :func:`compute_seed_tts_wer_metrics`). @@ -308,9 +308,9 @@ def build_arg_parser() -> argparse.ArgumentParser: p.add_argument( "--min-daily-omni-accuracy", type=float, - default=float((os.environ.get("ACC_BENCH_MIN_DAILY_OMNI_ACCURACY") or "0.69").strip() or "0.69"), + default=float((os.environ.get("ACC_BENCH_MIN_DAILY_OMNI_ACCURACY") or "0.67").strip() or "0.67"), help="Fail when daily_omni_accuracy is strictly below this threshold (0–1). " - "Default baseline 0.69; override with env ACC_BENCH_MIN_DAILY_OMNI_ACCURACY or pass 0 to disable the floor.", + "Default baseline 0.67; override with env ACC_BENCH_MIN_DAILY_OMNI_ACCURACY or pass 0 to disable the floor.", ) p.add_argument(