[router][ci] Add gpu utilization analyze with nvml (#10345)

2025-09-11 19:26:02 -07:00
parent 7bc5fb0d78
commit 7b141f816c
4 changed files with 329 additions and 35 deletions
--- a/sgl-router/py_test/e2e/test_regular_router.py
+++ b/sgl-router/py_test/e2e/test_regular_router.py
@@ -47,6 +47,8 @@ def test_genai_bench(
            "e2e_latency_mean_max": 14,
            "input_throughput_mean_min": 1000,
            "output_throughput_mean_min": 12,
+            # Enforce GPU utilization p50 >= 99% during the run.
+            "gpu_util_p50_min": 99,
        },
        kill_procs=e2e_two_workers_dp2,
    )