chore: update CI test est_time from recent run data

test/registered/attention/test_aiter_gluon_h12_fp8.py

@@ -15,7 +15,7 @@ from sglang.srt.layers.attention import aiter_mla_gluon as mod
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=15, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 _GLUON_FN = "sglang.srt.layers.attention.aiter_mla_gluon._gluon_fn"
 

test/registered/attention/test_chunk_gated_delta_rule.py

@@ -13,7 +13,7 @@ from sglang.test.ci.ci_register import (
     register_xpu_ci,
 )
 
-register_cuda_ci(est_time=6, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=7, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=11, stage="stage-b", runner_config="1-gpu-large-amd")
 register_xpu_ci(est_time=900, suite="stage-b-test-1-gpu-xpu")
 

test/registered/attention/test_create_kvindices.py

@@ -9,7 +9,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 # Triton kernel unit test for KV indices creation
-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/attention/test_deterministic.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
     is_in_amd_ci,
 )
 
-register_cuda_ci(est_time=280, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=272, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=278, suite="stage-b-test-1-gpu-small-amd")
 register_xpu_ci(est_time=207, suite="stage-b-test-1-gpu-xpu")
 

test/registered/attention/test_flash_attention_4.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashAttention4 integration test (requires SM 100+ / Blackwell B200)
-register_cuda_ci(est_time=230, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=220, stage="base-b", runner_config="4-gpu-b200")
 
 
 @unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")

test/registered/attention/test_hybrid_attn_backend.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
 
 # Hybrid attention backend tests (FA3 prefill + FlashInfer decode, requires SM 90+ / H100)
 # Multiple test classes: base, MLA, TorchCompile, SpecDecode variants
-register_cuda_ci(est_time=393, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=368, stage="extra-a", runner_config="1-gpu-large")
 
 
 class TestHybridAttnBackendMLA(TestHybridAttnBackendBase):

test/registered/attention/test_kda_kernels.py

@@ -19,7 +19,7 @@ from sglang.srt.utils.common import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=25, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=32, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=12, stage="stage-b", runner_config="1-gpu-large-amd")
 
 

test/registered/attention/test_normal_decode_set_metadata.py

@@ -21,7 +21,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 # Register this test for CUDA CI in base-b (fast attention/kernel tests)
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=17, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/test_qwen35_deterministic.py

@@ -12,7 +12,7 @@ from sglang.test.test_deterministic_utils import (
     TestDeterministicBase,
 )
 
-register_cuda_ci(est_time=119, stage="extra-b", runner_config="4-gpu-h100")
+register_cuda_ci(est_time=135, stage="extra-b", runner_config="4-gpu-h100")
 
 QWEN35 = "Qwen/Qwen3.5-35B-A3B"
 

test/registered/attention/test_torch_native_attention_backend.py

@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
 )
 
 # Torch native attention backend integration test with MMLU eval
-register_cuda_ci(est_time=310, stage="extra-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=312, stage="extra-a", runner_config="1-gpu-small")
 register_amd_ci(est_time=150, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/attention/test_triton_sliding_window.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 )
 
 # Sliding window attention with Triton backend (Gemma-3 model)
-register_cuda_ci(est_time=80, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=81, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/attention/test_trtllm_mha_encoder_only.py

@@ -17,7 +17,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 # trtllm_mha kernels are sm100-only; run this kernel-unit test on Blackwell.
-register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=8, stage="base-b", runner_config="4-gpu-b200")
 
 DEVICE = "cuda"
 PAGE_SIZE = 32

test/registered/attention/test_trtllm_mha_graph_metadata.py

@@ -23,7 +23,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardMode
 from sglang.test.ci.ci_register import register_cuda_ci
 
 # trtllm_mha kernels are sm100-only; run this kernel-unit test on Blackwell.
-register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
 
 DEVICE = "cuda"
 PAGE_SIZE = 128

test/registered/attention/test_trtllm_mha_page_table.py

@@ -25,7 +25,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 # Triton kernel unit test for the trtllm_mha device-side page-table build.
-register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=14, stage="stage-b", runner_config="1-gpu-small-amd")
 
 

test/registered/attention/test_unified_memory_deterministic.py

@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=71, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=90, stage="base-b", runner_config="1-gpu-large")
 
 KIMI_LINEAR_MODEL = "yujiepan/kimi-linear-tiny-random"
 SERVER_ENV = {"SGLANG_BATCH_INVARIANT_OPS_ENABLE_MM_DEEPGEMM": "0"}

test/registered/attention/unittests/dense/test_extend_init_contract.py

@@ -36,7 +36,7 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=11, stage="base-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-a", runner_config="1-gpu-small")
 
 _EXTEND_CASE = DenseAttentionCase(
     name="extend_no_prefix_smoke",

test/registered/attention/unittests/dense/test_fa3.py

@@ -30,7 +30,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=18, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/dense/test_fa4.py

@@ -29,8 +29,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=27, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=23, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=22, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=15, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/dense/test_flashinfer.py

@@ -26,8 +26,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=17, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=26, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(

test/registered/attention/unittests/dense/test_flex_attention.py

@@ -14,8 +14,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=17, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=36, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/dense/test_hybrid_attn.py

@@ -18,8 +18,8 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(

test/registered/attention/unittests/dense/test_tbo.py

@@ -21,8 +21,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/dense/test_torch_native.py

@@ -11,8 +11,8 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/dense/test_triton.py

@@ -30,8 +30,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=38, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=19, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=36, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=25, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/dense/test_trtllm_mha.py

@@ -28,8 +28,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=17, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=15, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=18, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(

test/registered/attention/unittests/dsa/test_dsa.py

@@ -30,8 +30,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_draft_runner i
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=14, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=22, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=26, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/dsv4/test_deepseek_v4.py

@@ -48,7 +48,7 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
 )
 
 register_cuda_ci(est_time=14, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/gdn/test_flashinfer.py

@@ -25,7 +25,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=15, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=14, stage="base-b", runner_config="4-gpu-b200")
 register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
 
 _cuda_major = int(torch.version.cuda.split(".")[0]) if torch.version.cuda else 0

test/registered/attention/unittests/gdn/test_linear_replayssm_decode.py

@@ -45,7 +45,7 @@ import torch
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/gdn/test_torch_native.py

@@ -15,7 +15,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
 register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 

test/registered/attention/unittests/gdn/test_triton.py

@@ -29,8 +29,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/hybrid_linear/test_flashinfer_mla_chunk_metadata.py

@@ -32,7 +32,7 @@ from sglang.test.kits.attention_unittest.attention_methods.mla_attention import
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 
 
 _KV_LORA_RANK = DEFAULT_KV_LORA_RANK

test/registered/attention/unittests/hybrid_linear/test_kda_fused_accept_indices.py

@@ -24,7 +24,7 @@ from sglang.srt.layers.attention.linear.kernels.kda_flashinfer import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-large")
 
 
 class TestBuildFusedAcceptIndices(CustomTestCase):

test/registered/attention/unittests/kda/test_triton.py

@@ -22,8 +22,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=15, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/lightning/test_triton.py

@@ -19,7 +19,7 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
 register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 

test/registered/attention/unittests/mamba/test_mamba2.py

@@ -32,8 +32,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/mamba/test_replay_state_indices_validator.py

@@ -7,7 +7,7 @@ from sglang.srt.layers.attention.mamba.replay_state_indices_validator import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=6, suite="base-a-test-cpu")
+register_cpu_ci(est_time=5, suite="base-a-test-cpu")
 
 
 class TestReplayStateIndicesValidator(unittest.TestCase):

test/registered/attention/unittests/mla/test_flashinfer.py

@@ -32,7 +32,7 @@ MLA_SHAPE_KWARGS = dict(
 
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
 register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 
 

test/registered/attention/unittests/mla/test_tokenspeed_mla.py

@@ -56,7 +56,7 @@ MLA_SHAPE_KWARGS = dict(
 from sglang.test.ci.ci_register import register_cuda_ci
 
 register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not _SUPPORTED, _SKIP_REASON)

test/registered/attention/unittests/mla/test_triton.py

@@ -28,8 +28,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=17, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")

test/registered/attention/unittests/swa/test_flashinfer.py

@@ -24,8 +24,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 
 
 @unittest.skipIf(

test/registered/attention/unittests/swa/test_swa_out_cache_loc.py

@@ -24,7 +24,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardMode
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=11, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class TestSWAKVPoolSetKVBuffer(CustomTestCase):

test/registered/attention/unittests/swa/test_torch_native.py

@@ -12,8 +12,8 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/attention/unittests/swa/test_triton.py

@@ -23,8 +23,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/backends/test_deepseek_v3_fp4_cutedsl_moe.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=243, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=259, stage="extra-b", runner_config="4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
 SERVER_LAUNCH_TIMEOUT = 1000

test/registered/backends/test_flashinfer_fusion_preflight.py

@@ -11,7 +11,7 @@ from sglang.srt.utils import get_cuda_driver_bindings, is_flashinfer_available
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=28, stage="base-b", runner_config="2-gpu-large")
+register_cuda_ci(est_time=26, stage="base-b", runner_config="2-gpu-large")
 
 WORLD_SIZE = 2
 

test/registered/backends/test_torch_compile.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=139, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=131, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/basic_perf/test_eagle3_latency.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=145, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=136, stage="extra-a", runner_config="1-gpu-large")
 
 
 class TestEagle3Latency(CustomTestCase):

test/registered/basic_perf/test_embeddings_api.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     run_embeddings_benchmark_multi,
 )
 
-register_cuda_ci(est_time=245, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=215, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=240, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/basic_perf/test_lora_latency.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=490, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=247, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=430, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/basic_perf/test_moe_throughput.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=290, stage="extra-a", runner_config="2-gpu-large")
+register_cuda_ci(est_time=271, stage="extra-a", runner_config="2-gpu-large")
 register_amd_ci(est_time=770, suite="stage-b-test-2-gpu-large-amd")
 
 

test/registered/basic_perf/test_pp_throughput.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=490, stage="extra-a", runner_config="2-gpu-large")
+register_cuda_ci(est_time=325, stage="extra-a", runner_config="2-gpu-large")
 register_amd_ci(est_time=1030, suite="stage-b-test-2-gpu-large-amd")
 
 

test/registered/basic_perf/test_score_api.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     run_score_benchmark_multi,
 )
 
-register_cuda_ci(est_time=215, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=185, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=210, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/basic_perf/test_serving_latency.py

@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=190, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=182, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=165, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/basic_perf/test_serving_throughput.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     run_bench_serving,
 )
 
-register_cuda_ci(est_time=710, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=660, stage="extra-a", runner_config="1-gpu-large")
 register_amd_ci(est_time=810, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/basic_perf/test_torch_compile_throughput.py

@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
     run_bench_offline_throughput,
 )
 
-register_cuda_ci(est_time=75, stage="extra-a", runner_config="2-gpu-large")
+register_cuda_ci(est_time=85, stage="extra-a", runner_config="2-gpu-large")
 register_amd_ci(est_time=280, suite="stage-b-test-2-gpu-large-amd")
 
 

test/registered/basic_perf/test_vlm_serving_fa3.py

@@ -6,7 +6,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.vlm_perf_kit import check_vlm_serving_perf
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=150, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=152, stage="extra-a", runner_config="1-gpu-large")
 
 
 class TestVLMServingFa3(CustomTestCase):

test/registered/basic_perf/test_vlm_serving_flashinfer.py

@@ -6,7 +6,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.vlm_perf_kit import check_vlm_serving_perf
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=195, stage="extra-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=191, stage="extra-a", runner_config="1-gpu-small")
 
 
 class TestVLMServingFlashinfer(CustomTestCase):

test/registered/bench_fn/test_bench_serving_reasoning_stream.py

@@ -26,7 +26,7 @@ from sglang.benchmark.serving import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=16, suite="base-a-test-cpu")
+register_cpu_ci(est_time=13, suite="base-a-test-cpu")
 
 
 def _free_port() -> int:

test/registered/bench_fn/test_benchmark_datasets_api.py

@@ -60,7 +60,7 @@ from sglang.benchmark.serving import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=38, suite="base-a-test-cpu")
+register_cpu_ci(est_time=36, suite="base-a-test-cpu")
 register_cpu_ci(est_time=46, suite="stage-b-test-cpu-intel")
 
 

test/registered/bench_fn/test_steady_state_benchmark.py

@@ -13,7 +13,7 @@ from sglang.benchmark.steady_state import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=10, suite="base-a-test-cpu")
+register_cpu_ci(est_time=7, suite="base-a-test-cpu")
 
 
 class _StringTokenizer:

test/registered/chunked_prefill/test_mm_chunked_embedding_unit.py

@@ -25,7 +25,7 @@ from sglang.srt.multimodal.transport.cuda_ipc import (
 from sglang.srt.runtime_context import get_context, get_parallel
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=12, suite="base-a-test-cpu")
+register_cpu_ci(est_time=9, suite="base-a-test-cpu")
 
 
 @pytest.fixture(autouse=True)

test/registered/chunked_prefill/test_scripted_core_1gpu.py

@@ -10,7 +10,7 @@ from sglang.test.scripted_runtime_chunked_helpers import (
     run_until_finished,
 )
 
-register_cuda_ci(est_time=106, stage="extra-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=103, stage="extra-a", runner_config="1-gpu-small")
 register_amd_ci(est_time=198, stage="extra-a", runner_config="1-gpu-small-amd")
 
 

test/registered/chunked_prefill/test_scripted_core_4gpu.py

@@ -11,7 +11,7 @@ from sglang.test.scripted_runtime_chunked_helpers import (
     run_until_finished,
 )
 
-register_cuda_ci(est_time=55, stage="extra-b", runner_config="4-gpu-h100")
+register_cuda_ci(est_time=96, stage="extra-b", runner_config="4-gpu-h100")
 
 
 _CHUNK_SIZE = 64

test/registered/chunked_prefill/test_scripted_swa_1gpu.py

@@ -5,7 +5,7 @@ from sglang.test.scripted_runtime.context import ScriptedContext
 from sglang.test.scripted_runtime.test_case import ScriptedTestCase
 from sglang.test.scripted_runtime_chunked_helpers import base_engine_kwargs
 
-register_cuda_ci(est_time=130, stage="extra-a", runner_config="1-gpu-large")
+register_cuda_ci(est_time=137, stage="extra-a", runner_config="1-gpu-large")
 
 
 _SWA_MODEL = "openai/gpt-oss-20b"

test/registered/constrained_decoding/test_constrained_decoding.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=161, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=149, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=220, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/core/test_basic_sanity.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=97, stage="base-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=112, stage="base-a", runner_config="1-gpu-small")
 register_amd_ci(est_time=160, suite="stage-a-test-1-gpu-small-amd")
 
 

test/registered/core/test_basic_sanity_dflash.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=135, stage="base-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=94, stage="base-a", runner_config="1-gpu-small")
 
 
 class TestBasicSanityDFlash(

test/registered/core/test_basic_sanity_dspark.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=97, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=92, stage="base-b", runner_config="1-gpu-large")
 
 TARGET_MODEL = "Qwen/Qwen3-14B"
 DRAFT_MODEL = "deepseek-ai/dspark_qwen3_14b_block7"

test/registered/core/test_basic_sanity_eagle3.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=167, stage="base-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=112, stage="base-a", runner_config="1-gpu-small")
 register_amd_ci(est_time=200, suite="stage-a-test-1-gpu-small-amd")
 
 

test/registered/core/test_engine_child_pids.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=77, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=34, stage="base-b", runner_config="1-gpu-small")
 
 
 class TestEngineChildPids(CustomTestCase):

test/registered/core/test_hidden_states.py

@@ -8,7 +8,7 @@ from sglang.srt.utils import get_device, is_hip
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST, CustomTestCase
 
-register_cuda_ci(est_time=32, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=31, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=55, suite="stage-b-test-1-gpu-small-amd")
 
 _is_hip = is_hip()

test/registered/core/test_no_extra_forked_cuda_context.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=54, stage="base-b", runner_config="2-gpu-large")
+register_cuda_ci(est_time=53, stage="base-b", runner_config="2-gpu-large")
 
 
 class TestTPServerGPUProcesses(CustomTestCase):

test/registered/core/test_request_queue_validation.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     send_generate_requests,
 )
 
-register_cuda_ci(est_time=53, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=63, stage="base-b", runner_config="1-gpu-small")
 
 
 class TestMaxQueuedRequests(CustomTestCase):

test/registered/core/test_srt_empty_deps.py

@@ -12,7 +12,7 @@ import pytest
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=7, suite="base-a-test-cpu")
+register_cpu_ci(est_time=5, suite="base-a-test-cpu")
 
 # Packages known to transitively depend on torch or triton.
 # If a new package is added to runtime_base and it pulls torch,

test/registered/core/test_srt_endpoint.py

@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
     run_logprob_check,
 )
 
-register_cuda_ci(est_time=265, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=250, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=260, suite="stage-b-test-1-gpu-small-amd")
 
 SERVER_ENV = {"SGLANG_USE_PICKLE_IPC": "0"}

test/registered/core/test_srt_engine.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=276, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=286, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=261, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/cp/test_cp_strategy_unit.py

@@ -41,7 +41,7 @@ from sglang.srt.runtime_context import get_parallel
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=11, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class _ExtendMode:

test/registered/cp/test_deepseek_v3_cp_single_node.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=275, stage="extra-b", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=334, stage="extra-b", runner_config="8-gpu-h200")
 
 DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 

test/registered/cp/test_deepseek_v4_flash_fp4_b200_cp.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=689, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=654, stage="extra-b", runner_config="4-gpu-b200")
 
 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/cp/test_dsa_prefill_cp.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=314, stage="extra-b", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=433, stage="extra-b", runner_config="8-gpu-h200")
 GLM52_MODEL_PATH = "zai-org/GLM-5.2-FP8"
 SERVER_LAUNCH_TIMEOUT = max(DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH, 1800)
 

test/registered/cp/test_gpt_oss_4gpu_mxfp4_cp.py

@@ -3,7 +3,7 @@ import unittest
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=121, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=127, stage="extra-b", runner_config="4-gpu-b200")
 
 
 class TestGptOss4GpuMxfp4CP(BaseTestGptOss):

test/registered/cp/test_gqa_prefill_cp.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=466, stage="extra-b", runner_config="4-gpu-h100")
+register_cuda_ci(est_time=553, stage="extra-b", runner_config="4-gpu-h100")
 
 GQA_MODEL_PATH = "Qwen/Qwen3-30B-A3B-FP8"
 

test/registered/cuda_graph/breakable/test_bcg_with_speculative_decoding.py

@@ -12,7 +12,7 @@ import unittest
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.server_fixtures.pcg_spec_fixture import PCGSpecBase
 
-register_cuda_ci(est_time=106, stage="base-b", runner_config="2-gpu-large")
+register_cuda_ci(est_time=103, stage="base-b", runner_config="2-gpu-large")
 
 
 class TestBCGWithEAGLE3(PCGSpecBase, unittest.TestCase):

test/registered/cuda_graph/breakable/test_breakable_cuda_graph.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
 )
 
 # CI Registration — large suite to fit the integration test's server startup.
-register_cuda_ci(est_time=90, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=88, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-c-test-large-8-gpu-amd-mi35x")
 
 

test/registered/cuda_graph/full_prefill/test_full_cuda_graph_prefill.py

@@ -33,7 +33,7 @@ from sglang.test.test_utils import (
 
 # OSS FA4 coverage requires Blackwell. The PP test uses two GPUs; the other
 # tests use one GPU.
-register_cuda_ci(est_time=251, stage="base-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=278, stage="base-b", runner_config="4-gpu-b200")
 
 
 def _prefill_graph_count(base_url: str) -> float:

test/registered/cuda_graph/test_cuda_piecewise_backend.py

@@ -5,7 +5,7 @@ import torch
 
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
 
 if not torch.cuda.is_available():
     pytest.skip("CUDA piecewise backend requires CUDA", allow_module_level=True)

test/registered/dcp/test_dcp_layout_unit.py

@@ -33,7 +33,7 @@ from sglang.srt.mem_cache.memory_pool import HybridLinearKVPool
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=11, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 DCP_SIZES = [1, 2, 3, 4, 8]
 LENS = list(range(0, 41))

test/registered/dcp/test_dsv31_dcp8_gsm8k.py

@@ -55,7 +55,7 @@ from sglang.test.test_utils import (
 # ---------------------------------------------------------------------------
 # CI registration — only TestDSV31DCP8TP8GSM8K runs in CI
 # ---------------------------------------------------------------------------
-register_cuda_ci(est_time=213, stage="extra-b", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=248, stage="extra-b", runner_config="8-gpu-h200")
 
 DEEPSEEK_V31_MODEL_PATH = "deepseek-ai/DeepSeek-V3.1"
 

test/registered/dcp/test_kimi_linear_dcp4.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=105, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=98, stage="extra-b", runner_config="4-gpu-b200")
 
 KIMI_LINEAR_MODEL = "moonshotai/Kimi-Linear-48B-A3B-Instruct"
 CUDA_GRAPH_MAX_BS_DECODE = 256

test/registered/dcp/test_kimi_linear_dcp_dspark4.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=201, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=206, stage="extra-b", runner_config="4-gpu-b200")
 
 KIMI_LINEAR_MODEL = "moonshotai/Kimi-Linear-48B-A3B-Instruct"
 GSM8K_SCORE_THRESHOLD = 0.88

test/registered/debug_utils/test_tensor_dump_forward_hook.py

@@ -18,7 +18,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.layer_ut_utils import init_single_process_dist
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
 # Backend-specific: the hook resolves get_device() and the matching distributed
 # backend, so only an AMD run exercises the HIP device and dump path.
 register_amd_ci(est_time=15, suite="stage-b-test-1-gpu-small-amd")

test/registered/disaggregation/test_disaggregation_aarch64.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 
 QWEN3_8B_MODEL_PATH = "Qwen/Qwen3-8B"
 
-register_cuda_ci(est_time=125, stage="base-c", runner_config="4-gpu-gb300")
+register_cuda_ci(est_time=137, stage="base-c", runner_config="4-gpu-gb300")
 
 # Keep rendezvous ports below the ephemeral range on the 4-GPU GB300 runner.
 NCCL_PORT_BASE = DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 120

test/registered/disaggregation/test_disaggregation_basic.py

@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     DEFAULT_TARGET_MODEL_EAGLE3,
 )
 
-register_cuda_ci(est_time=1007, stage="base-b", runner_config="2-gpu-large")
+register_cuda_ci(est_time=895, stage="base-b", runner_config="2-gpu-large")
 
 
 class TestDisaggregationAccuracy(PauseResumeInPlaceMixin, PDDisaggregationServerBase):

test/registered/disaggregation/test_disaggregation_chunked_prefill_abort.py

@@ -12,7 +12,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
 )
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 
-register_cuda_ci(est_time=120, stage="base-b", runner_config="2-gpu-large")
+register_cuda_ci(est_time=98, stage="base-b", runner_config="2-gpu-large")
 
 PD_CHUNKED_ABORT_EXTRA_ARGS = [
     "--max-running-requests",

test/registered/disaggregation/test_disaggregation_decode_radix_cache.py

@@ -26,7 +26,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=170, stage="base-c", runner_config="8-gpu-h20")
+register_cuda_ci(est_time=164, stage="base-c", runner_config="8-gpu-h20")
 
 
 def _has_mooncake():

test/registered/disaggregation/test_disaggregation_decode_radix_cache_swa.py

@@ -17,7 +17,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
 )
 from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST_MXFP4_WITH_MOE, is_in_ci
 
-register_cuda_ci(est_time=365, stage="extra-b", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=334, stage="extra-b", runner_config="8-gpu-h200")
 
 SWA_SERVER_ARGS = ["--page-size", "64", "--attention-backend", "triton"]
 

test/registered/disaggregation/test_disaggregation_different_tp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=1171, stage="base-c", runner_config="8-gpu-h20")
+register_cuda_ci(est_time=1180, stage="base-c", runner_config="8-gpu-h20")
 
 
 class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):

test/registered/disaggregation/test_disaggregation_dp_attention.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=140, stage="base-c", runner_config="8-gpu-h20")
+register_cuda_ci(est_time=141, stage="base-c", runner_config="8-gpu-h20")
 
 
 class TestDisaggregationDPAttention(PDDisaggregationServerBase):

test/registered/disaggregation/test_disaggregation_dsv4.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=184, stage="base-c", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=206, stage="base-c", runner_config="8-gpu-h200")
 
 DSV4_FLASH_MODEL = "sgl-project/DeepSeek-V4-Flash-FP8"