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

test/registered/models/test_compressed_tensors_models.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=65, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=65, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=42, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/models/test_dummy_grok_models.py

@@ -5,7 +5,8 @@ from sglang.test.test_utils import CustomTestCase, is_in_ci, run_bench_one_batch
 
 register_cuda_ci(
     est_time=120,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
     disabled="Temporarily disabled",
 )
 

test/registered/models/test_gemma4_fp8_per_expert_loading.py

@@ -25,7 +25,7 @@ from sglang.test.test_utils import (
 
 # Compressed-tensors per-expert FP8 MoE checkpoint that exercises the
 # loader path (gated repo + ~27 GB download + 4 GPUs at TP=4).
-register_cuda_ci(est_time=120, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=120, stage="stage-c", runner_config="4-gpu-h100")
 
 
 @unittest.skipIf(get_device_sm() < 90, "Test requires CUDA SM 90 or higher")

test/registered/models/test_generation_models.py

@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Generation model tests (CUDA only)
-register_cuda_ci(est_time=150, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=150, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=106, suite="stage-b-test-1-gpu-small-amd")
 
 # Copyright 2023-2024 SGLang Team

test/registered/models/test_kimi_linear_models.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=178, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=178, stage="stage-b", runner_config="2-gpu-large")
 
 
 class TestKimiLinear(CustomTestCase):

test/registered/models/test_ministral3_models.py

@@ -8,7 +8,8 @@ from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 
 register_cuda_ci(
     est_time=200,
-    suite="stage-b-test-1-gpu-small",
+    stage="stage-b",
+    runner_config="1-gpu-small",
     disabled="Temporarily disabled",
 )
 

test/registered/models/test_ministral4_models.py

@@ -8,7 +8,8 @@ from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 
 register_cuda_ci(
     est_time=200,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
 )
 
 MODEL = "mistralai/Mistral-Small-4-119B-2603"

test/registered/models/test_nvidia_nemotron_3_nano.py

@@ -6,7 +6,8 @@ from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 register_cuda_ci(
     est_time=190,
-    suite="stage-b-test-2-gpu-large",
+    stage="stage-b",
+    runner_config="2-gpu-large",
 )
 
 NEMOTRON_3_NANO_THINKING_ARGS = [

test/registered/models/test_transformers_backend_eval.py

@@ -7,7 +7,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=48, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=48, stage="stage-b", runner_config="1-gpu-small")
 
 
 class TestTransformersBackendEval(DefaultServerBase):

test/registered/models/test_transformers_models.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=177, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=177, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=320, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/models/test_vlm_models.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import is_in_ci
 # VLM (Vision Language Model) tests
 
 
-register_cuda_ci(est_time=317, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=317, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=850, suite="stage-b-test-1-gpu-small-amd-nondeterministic")
 
 _is_hip = is_hip()