ci: H200 conditional split + dsv4 est_time recalibration (h200 partition 6→2) (#25236)

test/registered/8-gpu-models/test_deepseek_v32_indexcache.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=492, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=450, suite="nightly-8-gpu-h200", nightly=True)
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 

test/registered/8-gpu-models/test_deepseek_v3_mtp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=309, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=300, stage="stage-c", runner_config="8-gpu-h200")
 
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 

test/registered/8-gpu-models/test_dsa_models_mtp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(
-    est_time=1048,
+    est_time=1030,
     stage="stage-c",
     runner_config="8-gpu-h200",
 )

test/registered/8-gpu-models/test_mimo_models.py

@@ -6,7 +6,7 @@ from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 
-register_cuda_ci(est_time=610, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=500, stage="stage-c", runner_config="8-gpu-h200")
 
 
 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):

test/registered/8-gpu-models/test_minimax_m25_basic.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=307, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=290, stage="stage-c", runner_config="8-gpu-h200")
 
 MINIMAX_M25_MODEL_PATH = "MiniMaxAI/MiniMax-M2.5"
 

test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=376, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=370, suite="nightly-8-gpu-h200", nightly=True)
 
 NEMOTRON_3_SUPER_BF16_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
 

test/registered/8-gpu-models/test_return_indexer_topk.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=270, suite="nightly-8-gpu-h200", nightly=True)
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 

test/registered/8-gpu-models/test_step3p5_flash_chain_mtp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=663, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=480, suite="nightly-8-gpu-h200", nightly=True)
 
 STEP3P5_FLASH_MODEL_PATH = "stepfun-ai/Step-3.5-Flash"
 

test/registered/distributed/test_disaggregation_hybrid_attention.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
-register_cuda_ci(est_time=695, stage="stage-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=310, suite="nightly-8-gpu-h200", nightly=True)
 
 
 @unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")

test/registered/dsv4/test_deepseek_v4_flash_fp4_b200.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=1800, stage="stage-c", runner_config="dsv4-4-gpu-b200")
+register_cuda_ci(est_time=700, stage="stage-c", runner_config="dsv4-4-gpu-b200")
 
 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/dsv4/test_deepseek_v4_flash_fp4_h200.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=1800, stage="stage-c", runner_config="dsv4-8-gpu-h200")
+register_cuda_ci(est_time=370, stage="stage-c", runner_config="dsv4-8-gpu-h200")
 
 
 def _flashinfer_has_sm90_cutlass_mxfp4() -> bool:

test/registered/dsv4/test_deepseek_v4_flash_fp8_h200.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=900, stage="stage-c", runner_config="dsv4-8-gpu-h200")
+register_cuda_ci(est_time=280, stage="stage-c", runner_config="dsv4-8-gpu-h200")
 
 MODEL_FP8 = "sgl-project/DeepSeek-V4-Flash-FP8"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/lora/test_lora_tp.py

@@ -30,7 +30,7 @@ from sglang.test.lora_utils import (
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
 register_cuda_ci(
-    est_time=190,
+    est_time=230,
     stage="stage-c",
     runner_config="8-gpu-h200",
 )