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

python/sglang/test/ci/ci_register.py

@@ -16,7 +16,12 @@ __all__ = [
     "ut_parse_one_file",
 ]
 
+# `suite` stays in positional slot 2 for backward compat with existing
+# `register_cpu_ci(5, "stage-a-test-cpu")` style positional calls. New fields
+# (`stage`, `runner_config`) are kwarg-only.
 _PARAM_ORDER = ("est_time", "suite", "nightly", "disabled")
+_KWARG_ONLY = ("stage", "runner_config")
+_ALL_PARAMS = _PARAM_ORDER + _KWARG_ONLY
 _UNSET = object()
 
 
@@ -31,25 +36,43 @@ class HWBackend(Enum):
 class CIRegistry:
     backend: HWBackend
     filename: str
-    # Estimated time to run the test in seconds.
     est_time: float
-    # The suite this test is registered in.
+    stage: Optional[str] = None
-    suite: str
+    runner_config: Optional[str] = None
-    # Whether the test is a nightly test.
+    # Legacy single-string suite; kept for nightly/stress/weekly + AMD/CPU/NPU
+    # suites whose names don't follow `{stage}-test-{runner_config}` shape.
+    suite: Optional[str] = None
     nightly: bool = False
-    # Reason for disabling the test. None = enabled, string = disabled with reason.
     disabled: Optional[str] = None
 
+    @property
+    def effective_suite(self) -> Optional[str]:
+        if self.stage is not None and self.runner_config is not None:
+            return f"{self.stage}-test-{self.runner_config}"
+        return self.suite
+
 
 def register_cpu_ci(
-    est_time: float, suite: str, nightly: bool = False, disabled: Optional[str] = None
+    est_time: float,
+    suite: Optional[str] = None,
+    nightly: bool = False,
+    disabled: Optional[str] = None,
+    *,
+    stage: Optional[str] = None,
+    runner_config: Optional[str] = None,
 ):
     """Marker for CPU CI registration (parsed via AST; runtime no-op)."""
     return None
 
 
 def register_cuda_ci(
-    est_time: float, suite: str, nightly: bool = False, disabled: Optional[str] = None
+    est_time: float,
+    suite: Optional[str] = None,
+    nightly: bool = False,
+    disabled: Optional[str] = None,
+    *,
+    stage: Optional[str] = None,
+    runner_config: Optional[str] = None,
 ):
     """Marker for CUDA CI registration (parsed via AST; runtime no-op)."""
     return None
@@ -57,9 +80,12 @@ def register_cuda_ci(
 
 def register_amd_ci(
     est_time: float,
-    suite: str,
+    suite: Optional[str] = None,
     nightly: bool = False,
     disabled: Optional[str] = None,
+    *,
+    stage: Optional[str] = None,
+    runner_config: Optional[str] = None,
 ):
     """Marker for AMD CI registration (parsed via AST; runtime no-op)."""
     return None
@@ -67,9 +93,12 @@ def register_amd_ci(
 
 def register_npu_ci(
     est_time: float,
-    suite: str,
+    suite: Optional[str] = None,
     nightly: bool = False,
     disabled: Optional[str] = None,
+    *,
+    stage: Optional[str] = None,
+    runner_config: Optional[str] = None,
 ):
     """Marker for NPU CI registration (parsed via AST; runtime no-op)."""
     return None
@@ -94,10 +123,8 @@ class RegistryVisitor(ast.NodeVisitor):
             return node.value
         return _UNSET
 
-    def _parse_call_args(
+    def _parse_call_args(self, func_call: ast.Call) -> dict:
-        self, func_call: ast.Call
+        args = {name: _UNSET for name in _ALL_PARAMS}
-    ) -> tuple[float, str, bool, Optional[str]]:
-        args = {name: _UNSET for name in _PARAM_ORDER}
         seen = set()
 
         if any(isinstance(arg, ast.Starred) for arg in func_call.args):
@@ -129,23 +156,49 @@ class RegistryVisitor(ast.NodeVisitor):
             seen.add(kw.arg)
             args[kw.arg] = self._constant_value(kw.value)
 
-        if args["est_time"] is _UNSET or args["suite"] is _UNSET:
+        if args["est_time"] is _UNSET:
             raise ValueError(
-                f"{self.filename}: est_time and suite are required constants in {func_call.func.id}()"
+                f"{self.filename}: est_time is a required constant in {func_call.func.id}()"
             )
 
-        est_time, suite = args["est_time"], args["suite"]
+        # The only valid (stage, runner_config, suite) shapes are:
-        nightly_value = args["nightly"]
+        #   (set,   set,   unset)  -> new-style pair
+        #   (unset, unset, set)    -> legacy single-string
+        # Any other combination is rejected with the actual triple in the error.
+        stage_set = args["stage"] is not _UNSET
+        runner_set = args["runner_config"] is not _UNSET
+        suite_set = args["suite"] is not _UNSET
+        valid_shape = (stage_set and runner_set and not suite_set) or (
+            not stage_set and not runner_set and suite_set
+        )
+        if not valid_shape:
+            raise ValueError(
+                f"{self.filename}: {func_call.func.id}() must specify exactly one of "
+                f"(stage, runner_config) pair or suite; got stage={stage_set}, "
+                f"runner_config={runner_set}, suite={suite_set}"
+            )
 
+        est_time = args["est_time"]
         if not isinstance(est_time, (int, float)):
             raise ValueError(
                 f"{self.filename}: est_time must be a number in {func_call.func.id}()"
             )
-        if not isinstance(suite, str):
+
+        suite = args["suite"] if suite_set else None
+        if suite is not None and not isinstance(suite, str):
             raise ValueError(
                 f"{self.filename}: suite must be a string in {func_call.func.id}()"
             )
 
+        stage = args["stage"] if stage_set else None
+        runner_config = args["runner_config"] if runner_set else None
+        for name, value in (("stage", stage), ("runner_config", runner_config)):
+            if value is not None and not isinstance(value, str):
+                raise ValueError(
+                    f"{self.filename}: {name} must be a string in {func_call.func.id}()"
+                )
+
+        nightly_value = args["nightly"]
         if nightly_value is _UNSET:
             nightly = False
         elif isinstance(nightly_value, bool):
@@ -161,7 +214,14 @@ class RegistryVisitor(ast.NodeVisitor):
                 f"{self.filename}: disabled must be a string in {func_call.func.id}()"
             )
 
-        return float(est_time), suite, nightly, disabled
+        return {
+            "est_time": float(est_time),
+            "stage": stage,
+            "runner_config": runner_config,
+            "suite": suite,
+            "nightly": nightly,
+            "disabled": disabled,
+        }
 
     def _collect_ci_registry(self, func_call: ast.Call):
         if not isinstance(func_call.func, ast.Name):
@@ -171,14 +231,11 @@ class RegistryVisitor(ast.NodeVisitor):
         if backend is None:
             return None
 
-        est_time, suite, nightly, disabled = self._parse_call_args(func_call)
+        parsed = self._parse_call_args(func_call)
         return CIRegistry(
             backend=backend,
             filename=self.filename,
-            est_time=est_time,
+            **parsed,
-            suite=suite,
-            nightly=nightly,
-            disabled=disabled,
         )
 
     @staticmethod

scripts/ci/utils/ci_coverage_report.py

@@ -155,7 +155,9 @@ def generate_summary_section(data: dict) -> str:
         for t in sorted(disabled_tests, key=lambda x: (x.backend.name, x.filename)):
             test_name = get_test_basename(t.filename)
             reason = t.disabled[:50] + "..." if len(t.disabled) > 50 else t.disabled
-            lines.append(f"| `{test_name}` | {t.backend.name} | {t.suite} | {reason} |")
+            lines.append(
+                f"| `{test_name}` | {t.backend.name} | {t.effective_suite} | {reason} |"
+            )
         lines.append("\n</details>\n")
 
     return "\n".join(lines)
@@ -197,7 +199,7 @@ def generate_by_folder_section(data: dict) -> str:
                     else ("Nightly" if t.nightly else "Per-Commit")
                 )
                 lines.append(
-                    f"| `{test_name}` | {t.suite} | {t.est_time:.0f}s | {status} |"
+                    f"| `{test_name}` | {t.effective_suite} | {t.est_time:.0f}s | {status} |"
                 )
 
             lines.append("")
@@ -231,7 +233,7 @@ def generate_by_suite_section(data: dict) -> str:
         # Group by suite within backend
         backend_suites = defaultdict(list)
         for t in backend_tests:
-            backend_suites[t.suite].append(t)
+            backend_suites[t.effective_suite].append(t)
 
         for suite in sorted(backend_suites.keys()):
             suite_tests = backend_suites[suite]
@@ -336,7 +338,7 @@ def generate_json_report(tests: list[CIRegistry]) -> str:
                 data["tests_by_folder"][folder]["backends"][backend] = [
                     {
                         "filename": get_test_basename(t.filename),
-                        "suite": t.suite,
+                        "suite": t.effective_suite,
                         "est_time": t.est_time,
                         "status": (
                             "disabled"
@@ -355,7 +357,7 @@ def generate_json_report(tests: list[CIRegistry]) -> str:
 
         backend_suites = defaultdict(list)
         for t in backend_tests:
-            backend_suites[t.suite].append(t)
+            backend_suites[t.effective_suite].append(t)
 
         data["tests_by_suite"][backend] = {
             "total": len(backend_tests),
@@ -423,7 +425,7 @@ def generate_json_report(tests: list[CIRegistry]) -> str:
             {
                 "filename": get_test_basename(t.filename),
                 "backend": t.backend.name,
-                "suite": t.suite,
+                "suite": t.effective_suite,
                 "reason": t.disabled,
             }
         )

scripts/ci/utils/compute_partitions.py

@@ -88,7 +88,7 @@ def compute_partitions(tests, full_parallel=False):
             continue
         if t.nightly or t.disabled is not None:
             continue
-        suite_tests[t.suite].append(t)
+        suite_tests[t.effective_suite].append(t)
 
     result = {}
     for suite, group in suite_tests.items():

test/registered/4-gpu-models/test_deepseek_v3_cutedsl_4gpu.py

@@ -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):

test/registered/4-gpu-models/test_gpt_oss_4gpu.py

@@ -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=392, stage="stage-c", runner_config="4-gpu-h100")
-register_cuda_ci(est_time=740, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=740, stage="stage-c", runner_config="4-gpu-b200")
 
 
 class TestGptOss4Gpu(BaseTestGptOss):

test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py

@@ -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"
 

test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py

@@ -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}}

test/registered/4-gpu-models/test_qwen35_hicache.py

@@ -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}}

test/registered/4-gpu-models/test_qwen35_models.py

@@ -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}}

test/registered/4-gpu-models/test_qwen3_30b.py

@@ -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"
 

test/registered/4-gpu-models/test_qwen3_next_models_mtp.py

@@ -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"
 

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, 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"
 

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, 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"
 

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

@@ -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"
 

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

@@ -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"

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, 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):

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, 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"
 

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, 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"
 

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, 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"
 

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, 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"
 

test/registered/attention/test_chunk_gated_delta_rule.py

@@ -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")

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=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")
 
 

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=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")

test/registered/attention/test_gdn_noncontiguous_stride.py

@@ -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"):

test/registered/attention/test_hybrid_attn_backend.py

@@ -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
 

test/registered/attention/test_kda_kernels.py

@@ -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(

test/registered/attention/test_normal_decode_set_metadata.py

@@ -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(

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=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")
 
 

test/registered/attention/test_triton_attention_backend.py

@@ -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")
 
 

test/registered/attention/test_triton_attention_kernels.py

@@ -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")
 
 

test/registered/attention/test_triton_sliding_window.py

@@ -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")
 
 

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=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")
 
 

test/registered/breakable_cuda_graph/test_breakable_cuda_graph.py

@@ -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):

test/registered/constrained_decoding/test_constrained_decoding.py

@@ -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")
 
 

test/registered/core/test_deterministic.py

@@ -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")
 
 

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

test/registered/core/test_gemma4_moe_deterministic.py

@@ -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 = (

test/registered/core/test_gpt_oss_sm120.py

@@ -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")

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=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()

test/registered/core/test_mm_process_config.py

@@ -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")
 
 

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, 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")
 
 

test/registered/core/test_srt_endpoint.py

@@ -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")
 
 

test/registered/core/test_srt_engine.py

@@ -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")
 
 

test/registered/cp/test_deepseek_v32_cp_single_node.py

@@ -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"
 

test/registered/debug_utils/test_cuda_coredump.py

@@ -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",
 )
 

test/registered/debug_utils/test_tensor_dump_forward_hook.py

@@ -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(

test/registered/disaggregation/test_disaggregation_basic.py

@@ -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):

test/registered/disaggregation/test_disaggregation_decode_offload.py

@@ -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.",
 )
 

test/registered/disaggregation/test_disaggregation_xpu.py

@@ -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",
 )
 

test/registered/disaggregation/test_specv2_kvcache_offloading.py

@@ -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(

test/registered/distributed/test_data_parallelism.py

@@ -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")
 
 

test/registered/distributed/test_disaggregation_aarch64.py

@@ -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):

test/registered/distributed/test_disaggregation_decode_radix_cache.py

@@ -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():

test/registered/distributed/test_disaggregation_different_tp.py

@@ -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):

test/registered/distributed/test_disaggregation_dp_attention.py

@@ -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):

test/registered/distributed/test_disaggregation_dsv4.py

@@ -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"
 

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, 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.")

test/registered/distributed/test_disaggregation_pp.py

@@ -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):

test/registered/distributed/test_dp_attention.py

@@ -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(

test/registered/distributed/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=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
 

test/registered/distributed/test_load_weights_from_remote_instance.py

@@ -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")
 
 

test/registered/distributed/test_parallel_state.py

@@ -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

test/registered/distributed/test_pp_single_node.py

@@ -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")
 
 

test/registered/dllm/test_llada2_mini.py

@@ -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

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, 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

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, 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:

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, 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

test/registered/ep/test_deepep_large.py

@@ -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"
 

test/registered/ep/test_deepep_small.py

@@ -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):

test/registered/ep/test_mooncake_ep_small.py

@@ -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()
 

test/registered/hicache/test_hicache_spec_file_storage.py

@@ -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.")

test/registered/hicache/test_hicache_storage.py

@@ -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

test/registered/hicache/test_hicache_storage_3fs_backend.py

@@ -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")
 
 

test/registered/hicache/test_hicache_storage_file_backend.py

@@ -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")
 
 

test/registered/hicache/test_hicache_storage_mooncake_backend.py

@@ -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):

test/registered/hicache/test_hicache_storage_runtime_attach_detach.py

@@ -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):

test/registered/hicache/test_hicache_variants.py

@@ -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.

test/registered/input_embedding/test_input_embeddings.py

@@ -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")
 
 

test/registered/input_embedding/test_input_embeds_chunked.py

@@ -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

test/registered/kernels/test_nsa_indexer.py

@@ -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 = {

test/registered/language/test_srt_backend.py

@@ -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")
 
 

test/registered/layers/mamba/test_causal_conv1d.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=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

test/registered/layers/mamba/test_mamba2_mixer.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
 

test/registered/layers/mamba/test_mamba_ssm.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=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

test/registered/layers/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

test/registered/layers/test_fla_layernorm_guard.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",
 )
 

test/registered/lora/test_fused_moe_lora_kernel.py

@@ -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):

test/registered/lora/test_lora_drainer.py

@@ -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

test/registered/lora/test_lora_eviction.py

@@ -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 = [

test/registered/lora/test_lora_gpt_oss_20b_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_hf_sgl_logprob_diff.py

@@ -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,

test/registered/lora/test_lora_moe_tp_logprob_diff.py

@@ -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

test/registered/lora/test_lora_moe_vllm_sgl_logprob_diff.py

@@ -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, ...]}, ...]

test/registered/lora/test_lora_nemotron_3_super_120b_a12b_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_overlap_loading.py

@@ -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")
 
 

test/registered/lora/test_lora_qwen3_30b_a3b_instruct_2507_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_qwen3_5_35b_a3b_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_qwen3_5_4b_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_qwen3_8b_logprob_diff.py

@@ -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"

test/registered/lora/test_lora_qwen3_vl_30b_a3b_instruct_logprob_diff.py

@@ -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"