[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_pd_router.py
+++ b/sgl-router/py_test/e2e/test_pd_router.py
@@ -257,6 +257,7 @@ def test_pd_genai_bench(e2e_model: str, pd_cluster, genai_bench_runner):
            "e2e_latency_mean_max": 15,
            "input_throughput_mean_min": 400,
            "output_throughput_mean_min": 20,
+            "gpu_util_p50_min": 99,
        },
        kill_procs=pd_cluster.workers,
    )