diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index a69202bb7..b00d0a297 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -19,7 +19,7 @@ suites = { "per-commit-8-gpu-h200": [ TestFile("test_deepseek_v3_basic.py", 275), TestFile("test_deepseek_v3_mtp.py", 275), - TestFile("test_disaggregation_hybrid_attention.py", 600), + TestFile("test_disaggregation_hybrid_attention.py", 400), TestFile("models/test_kimi_k2_models.py", 200), TestFile("test_deepseek_v32_basic.py", 360), TestFile("test_deepseek_v32_mtp.py", 360), diff --git a/test/srt/test_disaggregation_hybrid_attention.py b/test/srt/test_disaggregation_hybrid_attention.py index cf310515c..276325982 100644 --- a/test/srt/test_disaggregation_hybrid_attention.py +++ b/test/srt/test_disaggregation_hybrid_attention.py @@ -7,10 +7,12 @@ from sglang.test.server_fixtures.disaggregation_fixture import ( ) from sglang.test.test_utils import ( DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH, + is_in_ci, popen_launch_pd_server, ) +@unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.") class TestDisaggregationHybridAttentionMamba(PDDisaggregationServerBase): @classmethod def setUpClass(cls):