From 96a73c4e15a2b98672b83b91af47d1e3bada707c Mon Sep 17 00:00:00 2001 From: Bingxu Chen Date: Wed, 26 Aug 2026 00:41:36 +0800 Subject: [PATCH] [AMD][CI] Adjust MI300 score API performance thresholds (#36290) --- .../registered/perf/test_bench_serving_1gpu_part2.py | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/test/registered/perf/test_bench_serving_1gpu_part2.py b/test/registered/perf/test_bench_serving_1gpu_part2.py index f0e242ec7..97c5f6f57 100644 --- a/test/registered/perf/test_bench_serving_1gpu_part2.py +++ b/test/registered/perf/test_bench_serving_1gpu_part2.py @@ -94,9 +94,15 @@ class TestBenchServing1GPUPart2(CustomTestCase): ) self.assertEqual(res["successful_requests"], res["total_requests"]) - self.assertLess(res["avg_latency_ms"], 48) - self.assertLess(res["p95_latency_ms"], 50) - self.assertGreater(res["throughput"], 20) + # relax for mi300x + if is_in_amd_ci(): + self.assertLess(res["avg_latency_ms"], 60) + self.assertLess(res["p95_latency_ms"], 65) + self.assertGreater(res["throughput"], 16) + else: + self.assertLess(res["avg_latency_ms"], 48) + self.assertLess(res["p95_latency_ms"], 50) + self.assertGreater(res["throughput"], 20) def test_score_api_batch_scaling(self): """Test score API performance with different batch sizes"""