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

test/registered/moe/test_cutedsl_moe.py

@@ -16,7 +16,7 @@ except ImportError:
     CuteDslMoEWrapper = None
     convert_sf_to_mma_layout = None
 
-register_cuda_ci(est_time=590, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=590, stage="stage-c", runner_config="4-gpu-b200")
 
 SKIP_TEST = torch.cuda.get_device_capability() < (10, 0)
 SKIP_REASON = "Nvfp4 Requires compute capability of 10 or above."

test/registered/moe/test_fused_moe.py

@@ -13,7 +13,7 @@ from sglang.srt.utils import get_device, get_device_capability, is_hip
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, empty_gpu_cache
 
-register_cuda_ci(est_time=87, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=87, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd")
 
 _is_hip = is_hip()

test/registered/moe/test_glm4_moe_models.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=171, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=171, stage="stage-b", runner_config="2-gpu-large")
 
 
 class TestGLM4MoE(CustomTestCase):

test/registered/moe/test_moe_ep.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=279, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=279, stage="stage-b", runner_config="2-gpu-large")
 
 
 class TestEp(CustomTestCase):

test/registered/moe/test_torch_compile_moe.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=130, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=130, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=1400, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/moe/test_triton_fused_moe.py

@@ -12,7 +12,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=13, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=13, stage="stage-b", runner_config="1-gpu-large")
 
 
 class TestFusedMOE(CustomTestCase):

test/registered/moe/test_triton_moe_channel_fp8_kernel.py

@@ -11,7 +11,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=17, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=17, stage="stage-b", runner_config="1-gpu-large")
 
 
 def native_w8a8_per_token_matmul(A, B, As, Bs, output_dtype=torch.float16):