ci: decouple stage and runner for cuda registry (#25197)

2026-05-13 17:28:21 -07:00
parent dd98153d17
commit 22d3f3996c
261 changed files with 388 additions and 293 deletions
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=1800, suite="stage-c-test-4-gpu-gb200")
+register_cuda_ci(est_time=1800, stage="stage-c", runner_config="4-gpu-gb200")


 class TestDeepseekR1Nvfp4CuteDSLDeepEP(CustomTestCase):
@@ -3,8 +3,8 @@ 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=392, suite="stage-c-test-4-gpu-h100")
-register_cuda_ci(est_time=740, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=392, stage="stage-c", runner_config="4-gpu-h100")
+register_cuda_ci(est_time=740, stage="stage-c", runner_config="4-gpu-b200")


 class TestGptOss4Gpu(BaseTestGptOss):
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=710, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=710, stage="stage-c", runner_config="4-gpu-b200")

 NEMOTRON_3_SUPER_NVFP4_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=540, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=540, stage="stage-c", runner_config="4-gpu-b200")

 QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
 ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=540, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=540, stage="stage-c", runner_config="4-gpu-h100")

 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 ACC_THRESHOLDS = {QWEN35_27B_MODEL: {"gsm8k": 0.8}}
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=260, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=260, stage="stage-c", runner_config="4-gpu-b200")

 QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
 ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=261, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=261, stage="stage-c", runner_config="4-gpu-h100")

 QWEN3_30B_MODEL_PATH = "Qwen/Qwen3-30B-A3B-FP8"

@@ -6,7 +6,7 @@ from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
 from sglang.test.kits.prefix_cache_branching_kit import PrefixCacheBranchingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase

-register_cuda_ci(est_time=290, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=290, stage="stage-c", runner_config="4-gpu-h100")

 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
    write_github_step_summary,
 )

-register_cuda_ci(est_time=492, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=492, stage="stage-c", runner_config="8-gpu-h200")

 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
    write_github_step_summary,
 )

-register_cuda_ci(est_time=309, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=309, stage="stage-c", runner_config="8-gpu-h200")

 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"

@@ -12,7 +12,9 @@ from sglang.test.test_utils import (
    write_github_step_summary,
 )

-register_cuda_ci(est_time=720, suite="stage-c-test-8-gpu-h200", nightly=True)
+register_cuda_ci(
+    est_time=720, stage="stage-c", runner_config="8-gpu-h200", nightly=True
+)

 GLM5_MODEL_PATH = "zai-org/GLM-5-FP8"

@@ -18,7 +18,8 @@ from sglang.test.test_utils import (

 register_cuda_ci(
    est_time=1048,
-    suite="stage-c-test-8-gpu-h200",
+    stage="stage-c",
+    runner_config="8-gpu-h200",
 )

 FULL_DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
@@ -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, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=610, stage="stage-c", runner_config="8-gpu-h200")


 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    write_github_step_summary,
 )

-register_cuda_ci(est_time=307, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=307, stage="stage-c", runner_config="8-gpu-h200")

 MINIMAX_M25_MODEL_PATH = "MiniMaxAI/MiniMax-M2.5"

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=376, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=376, stage="stage-c", runner_config="8-gpu-h200")

 NEMOTRON_3_SUPER_BF16_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=600, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=600, stage="stage-c", runner_config="8-gpu-h200")

 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
    write_github_step_summary,
 )

-register_cuda_ci(est_time=663, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=663, stage="stage-c", runner_config="8-gpu-h200")

 STEP3P5_FLASH_MODEL_PATH = "stepfun-ai/Step-3.5-Flash"

@@ -8,7 +8,7 @@ from sglang.srt.layers.attention.fla.fused_recurrent import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, stage="stage-b", runner_config="1-gpu-large")


@unittest.skipIf(not torch.cuda.is_available(), "Test requires CUDA")
@@ -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=7, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=7, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")


@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )

 # FlashAttention4 integration test (requires SM 100+ / Blackwell B200)
-register_cuda_ci(est_time=265, suite="stage-b-test-4-gpu-b200")
+register_cuda_ci(est_time=265, stage="stage-b", runner_config="4-gpu-b200")


@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
@@ -14,7 +14,7 @@ from sglang.srt.layers.attention.fla.fused_sigmoid_gating_recurrent import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=7, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=7, stage="stage-b", runner_config="1-gpu-large")


 def _make_noncontiguous_ab(batch, num_heads, dtype=torch.bfloat16, device="cuda"):
@@ -20,7 +20,7 @@ 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=407, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=407, stage="stage-b", runner_config="1-gpu-large")

 GSM_DATASET_PATH = None

@@ -14,7 +14,7 @@ from sglang.srt.layers.attention.fla.kda import (
 from sglang.srt.utils.common import get_device
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=12, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=12, stage="stage-b", runner_config="1-gpu-large")


@unittest.skipIf(
@@ -21,7 +21,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase

 # Register this test for CUDA CI in stage-b (fast attention/kernel tests)
-register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, stage="stage-b", runner_config="1-gpu-large")


 def reference_normal_decode_set_metadata(
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
 )

 # Torch native attention backend integration test with MMLU eval
-register_cuda_ci(est_time=140, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=140, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=150, suite="stage-b-test-1-gpu-small-amd")


@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
 )

 # Triton attention backend integration test with latency benchmark and MMLU eval
-register_cuda_ci(est_time=177, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=177, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=1400, suite="stage-b-test-1-gpu-small-amd")


@@ -23,7 +23,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, is_in_amd_ci

 # Triton attention kernel unit tests (decode, extend, prefill)
-register_cuda_ci(est_time=19, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=19, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd")


@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )

 # Sliding window attention with Triton backend (Gemma-3 model)
-register_cuda_ci(est_time=93, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=93, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-small-amd")


@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=126, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=126, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-small-amd")


@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
 )

 # CI Registration — large suite to fit the integration test's server startup.
-register_cuda_ci(est_time=79, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=79, stage="stage-b", runner_config="1-gpu-large")


 def _skip_if_no_cuda(test_func):
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=120, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=179, suite="stage-b-test-1-gpu-small-amd")


@@ -16,7 +16,7 @@ from sglang.test.test_deterministic_utils import (
 )
 from sglang.test.test_utils import is_in_amd_ci

-register_cuda_ci(est_time=207, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=207, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=278, suite="stage-b-test-1-gpu-small-amd")


@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
    CustomTestCase,
 )

-register_cuda_ci(est_time=77, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=77, stage="stage-b", runner_config="1-gpu-small")


 class TestEngineChildPids(CustomTestCase):
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=107, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=107, stage="stage-b", runner_config="2-gpu-large")


 PROMPT = (
@@ -5,7 +5,7 @@ import torch
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss

-register_cuda_ci(est_time=345, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=345, stage="stage-b", runner_config="1-gpu-small")


@unittest.skipIf(not torch.cuda.is_available(), "CUDA is not available")
@@ -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=45, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=45, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=55, suite="stage-b-test-1-gpu-small-amd")

 _is_hip = is_hip()
@@ -4,7 +4,7 @@ from unittest.mock import MagicMock, patch
 from sglang.srt.server_args import ServerArgs
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=1, suite="stage-b-test-1-gpu-small-amd")


@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
    send_generate_requests,
 )

-register_cuda_ci(est_time=53, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=53, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=70, suite="stage-b-test-1-gpu-small-amd")


@@ -28,7 +28,7 @@ from sglang.test.test_utils import (
    run_logprob_check,
 )

-register_cuda_ci(est_time=134, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=134, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=130, suite="stage-b-test-1-gpu-small-amd")


@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
    CustomTestCase,
 )

-register_cuda_ci(est_time=387, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=387, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=261, suite="stage-b-test-1-gpu-small-amd")


@@ -15,7 +15,8 @@ from sglang.test.test_utils import (

 register_cuda_ci(
    est_time=616,
-    suite="stage-c-test-deepep-8-gpu-h200",
+    stage="stage-c",
+    runner_config="deepep-8-gpu-h200",
 )
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"

@@ -12,7 +12,8 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=10,
-    suite="stage-a-test-1-gpu-small",
+    stage="stage-a",
+    runner_config="1-gpu-small",
    disabled="Manual only: triggers intentional CUDA crash for coredump verification",
 )

@@ -19,7 +19,8 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

 register_cuda_ci(
    est_time=9,
-    suite="stage-b-test-1-gpu-small",
+    stage="stage-b",
+    runner_config="1-gpu-small",
    disabled="Test uses pytest-style function without TestCase class - see #17145",
 )
 register_amd_ci(
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    DEFAULT_TARGET_MODEL_EAGLE3,
 )

-register_cuda_ci(est_time=509, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=509, stage="stage-b", runner_config="2-gpu-large")


 class TestDisaggregationAccuracy(PauseResumeInPlaceMixin, PDDisaggregationServerBase):
@@ -19,7 +19,8 @@ from sglang.test.test_utils import (
 # Increasing estimated time since we run evaluation twice
 register_cuda_ci(
    est_time=600,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
    disabled="Temporarily disable the flaky test.",
 )

@@ -26,7 +26,8 @@ from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST_QWEN

 register_cuda_ci(
    est_time=300,
-    suite="stage-a-test-1-gpu-small",
+    stage="stage-a",
+    runner_config="1-gpu-small",
    disabled="Intel XPU only — not available in standard CUDA CI",
 )

@@ -17,7 +17,7 @@ from sglang.srt.disaggregation.decode_kvcache_offload_manager import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=8, stage="stage-b", runner_config="1-gpu-small")


 def _make_mock_req(
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=91, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=91, stage="stage-b", runner_config="2-gpu-large")
 register_amd_ci(est_time=73, suite="stage-b-test-2-gpu-large-amd")


@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
    popen_launch_pd_server,
 )

-register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-gb200")
+register_cuda_ci(est_time=300, stage="stage-c", runner_config="4-gpu-gb200")


 class TestDisaggregationMooncakeAARCH64Accuracy(PDDisaggregationServerBase):
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=300, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=300, stage="stage-c", runner_config="8-gpu-h20")


 def _has_nixl():
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=375, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=375, stage="stage-c", runner_config="8-gpu-h20")


 class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=443, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=443, stage="stage-c", runner_config="8-gpu-h20")


 class TestDisaggregationDPAttention(PDDisaggregationServerBase):
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=250, suite="stage-c-test-dsv4-8-gpu-h200")
+register_cuda_ci(est_time=250, stage="stage-c", runner_config="dsv4-8-gpu-h200")

 DSV4_FLASH_MODEL = "sgl-project/DeepSeek-V4-Flash-FP8"

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
    popen_launch_pd_server,
 )

-register_cuda_ci(est_time=695, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=695, stage="stage-c", runner_config="8-gpu-h200")


@unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=216, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=216, stage="stage-c", runner_config="8-gpu-h20")


 class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=420, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=420, stage="stage-b", runner_config="2-gpu-large")


 class TestDPAttentionDP2TP2(
@@ -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=30, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=30, stage="stage-b", runner_config="2-gpu-large")

 WORLD_SIZE = 2

@@ -38,7 +38,7 @@ from sglang.utils import terminate_process

 mp.set_start_method("spawn", force=True)

-register_cuda_ci(est_time=145, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=145, stage="stage-b", runner_config="2-gpu-large")
 register_amd_ci(est_time=72, suite="stage-b-test-2-gpu-large-amd")


@@ -43,7 +43,7 @@ import pytest

 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=8, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=8, suite="stage-b-test-1-gpu-small-amd")

 # Import the actual parallel_state module
@@ -32,7 +32,7 @@ from sglang.test.test_utils import (
    run_bench_one_batch_server,
 )

-register_cuda_ci(est_time=554, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=554, stage="stage-c", runner_config="4-gpu-h100")
 register_amd_ci(est_time=650, suite="stage-c-test-4-gpu-amd")


@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=139, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=139, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=330, suite="stage-b-test-1-gpu-small-amd")

 import unittest
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=1800, suite="stage-c-test-dsv4-4-gpu-b200")
+register_cuda_ci(est_time=1800, stage="stage-c", runner_config="dsv4-4-gpu-b200")

 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=1800, suite="stage-c-test-dsv4-8-gpu-h200")
+register_cuda_ci(est_time=1800, stage="stage-c", runner_config="dsv4-8-gpu-h200")


 def _flashinfer_has_sm90_cutlass_mxfp4() -> bool:
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
    try_cached_model,
 )

-register_cuda_ci(est_time=900, suite="stage-c-test-dsv4-8-gpu-h200")
+register_cuda_ci(est_time=900, stage="stage-c", runner_config="dsv4-8-gpu-h200")

 MODEL_FP8 = "sgl-project/DeepSeek-V4-Flash-FP8"
 SERVER_LAUNCH_TIMEOUT = 3600
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=528, suite="stage-c-test-deepep-8-gpu-h200")
+register_cuda_ci(est_time=528, stage="stage-c", runner_config="deepep-8-gpu-h200")

 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=478, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=478, stage="stage-c", runner_config="deepep-4-gpu-h100")


 class TestPureDP(CustomTestCase):
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=82, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=82, stage="stage-c", runner_config="deepep-4-gpu-h100")

 ib_devices = get_rdma_devices_args()

@@ -30,7 +30,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import wait_for_http_ready

-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=200, stage="stage-b", runner_config="1-gpu-large")


@unittest.skipIf(is_hip(), "HiCache + EAGLE3 file-storage loadback e2e is CUDA-only.")
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=99, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=99, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=300, suite="stage-b-test-1-gpu-small-amd")

 import time
@@ -13,7 +13,7 @@ from test_hicache_storage_file_backend import HiCacheStorageBaseMixin
 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=150, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=150, stage="stage-b", runner_config="2-gpu-large")
 register_amd_ci(est_time=300, suite="stage-b-test-2-gpu-large")


@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import wait_for_http_ready

-register_cuda_ci(est_time=148, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=148, stage="stage-b", runner_config="2-gpu-large")
 register_amd_ci(est_time=526, suite="stage-b-test-2-gpu-large-amd")


@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    is_in_ci,
 )

-register_cuda_ci(est_time=236, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=236, stage="stage-b", runner_config="2-gpu-large")


 class HiCacheStorageMooncakeBackendBaseMixin(HiCacheStorageBaseMixin):
@@ -28,7 +28,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import wait_for_http_ready

-register_cuda_ci(est_time=139, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=139, stage="stage-b", runner_config="2-gpu-large")


 class TestHiCacheStorageRuntimeAttachDetach(CustomTestCase):
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=450, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=450, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=524, suite="stage-b-test-1-gpu-small-amd")
 """
 Consolidated HiCache variant tests.
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=42, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=38, suite="stage-b-test-1-gpu-small-amd")


@@ -30,7 +30,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cuda_ci(est_time=43, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=43, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=43, suite="stage-b-test-1-gpu-small-amd")

 CHUNKED_PREFILL_SIZE = 256
@@ -24,7 +24,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMo
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.test.test_utils import CustomTestCase

-register_cuda_ci(est_time=18, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=18, stage="stage-b", runner_config="1-gpu-large")

 # Global configuration for all indexer tests
 DEFAULT_CONFIG = {
@@ -20,7 +20,7 @@ from sglang.test.test_programs import (
 )
 from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST, CustomTestCase

-register_cuda_ci(est_time=79, suite="stage-a-test-1-gpu-small")
+register_cuda_ci(est_time=79, stage="stage-a", runner_config="1-gpu-small")
 register_amd_ci(est_time=120, suite="stage-a-test-1-gpu-small-amd")


@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=25, suite="stage-b-test-1-gpu-small-amd")

 # Adapted from https://github.com/vllm-project/vllm/blob/main/tests/kernels/mamba/test_causal_conv1d.py
@@ -18,7 +18,7 @@ from sglang.srt.distributed.parallel_state import (
 from sglang.srt.utils import get_device, get_device_count
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=32, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=32, stage="stage-b", runner_config="2-gpu-large")

 NUM_GPUS = 2

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-small-amd")

 # Adapted from https://github.com/vllm-project/vllm/blob/633f943e30a4444d890d26b81850f7217736f840/tests/kernels/mamba/test_mamba_ssm_ssd.py
@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=34, suite="stage-b-test-1-gpu-small-amd")

 # Adapted from https://github.com/vllm-project/vllm/blob/633f943e30a4444d890d26b81850f7217736f840/tests/kernels/mamba/test_mamba_ssm_ssd.py
@@ -19,7 +19,8 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=60,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
    disabled="Temporarily disabled",
 )

@@ -15,7 +15,7 @@ from sglang.test.ci.ci_register import register_cuda_ci

 # ==============================================================================

-register_cuda_ci(est_time=28, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=28, stage="stage-b", runner_config="1-gpu-large")


 def round_up(x, base):
@@ -12,7 +12,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import is_in_ci

-register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=100, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")

 MOCK_START_TIME = 1000.0
@@ -23,7 +23,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.runners import SRTRunner
 from sglang.test.test_utils import CustomTestCase

-register_cuda_ci(est_time=263, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=263, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=224, suite="stage-b-test-1-gpu-small-amd")

 PROMPTS = [
@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=300,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )

 BASE_MODEL = "lmsys/gpt-oss-20b-bf16"
@@ -41,7 +41,8 @@ from sglang.test.test_utils import DEFAULT_PORT_FOR_SRT_TEST_RUNNER, CustomTestC

 register_cuda_ci(
    est_time=150,
-    suite="stage-b-test-1-gpu-small",
+    stage="stage-b",
+    runner_config="1-gpu-small",
 )
 register_amd_ci(
    est_time=250,
@@ -34,7 +34,8 @@ from sglang.test.test_utils import (

 register_cuda_ci(
    est_time=200,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
 )

 LOGPROB_THRESHOLD = 5e-04
@@ -26,7 +26,8 @@ from sglang.test.runners import SRTRunner

 register_cuda_ci(
    est_time=50,
-    suite="stage-b-test-1-gpu-large",
+    stage="stage-b",
+    runner_config="1-gpu-large",
 )

 # Format: [{"text": "result string", "lps": [0.1, 0.2, ...]}, ...]
@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=300,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )

 BASE_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
@@ -29,7 +29,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cuda_ci(est_time=48, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=48, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=75, suite="stage-b-test-1-gpu-small-amd")


@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=160,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )

 BASE_MODEL = "Qwen/Qwen3-30B-A3B-Instruct-2507"
@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=160,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )

 BASE_MODEL = "Qwen/Qwen3.5-35B-A3B"
@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=90,
-    suite="stage-b-test-1-gpu-large",
+    stage="stage-b",
+    runner_config="1-gpu-large",
 )

 BASE_MODEL = "Qwen/Qwen3.5-4B"
@@ -39,7 +39,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=40,
-    suite="stage-b-test-1-gpu-large",
+    stage="stage-b",
+    runner_config="1-gpu-large",
 )

 BASE_MODEL = "Qwen/Qwen3-8B"
@@ -36,7 +36,8 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(
    est_time=160,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )

 BASE_MODEL = "Qwen/Qwen3-VL-30B-A3B-Instruct"
@@ -31,7 +31,8 @@ from sglang.test.test_utils import CustomTestCase, is_in_ci

 register_cuda_ci(
    est_time=190,
-    suite="stage-c-test-8-gpu-h200",
+    stage="stage-c",
+    runner_config="8-gpu-h200",
 )
 register_amd_ci(
    est_time=116,
@@ -36,7 +36,8 @@ from sglang.test.test_utils import (

 register_cuda_ci(
    est_time=487,
-    suite="stage-b-test-1-gpu-large",
+    stage="stage-b",
+    runner_config="1-gpu-large",
 )

 PROMPTS = [
@@ -25,7 +25,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase, is_in_ci

-register_cuda_ci(est_time=99, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=99, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")


--- a/Show More
+++ b/Show More